dev
This commit is contained in:
echo840
150
scripts/GPT4V.py
Normal file
150
scripts/GPT4V.py
Normal file
@@ -0,0 +1,150 @@
|
||||
import base64
|
||||
import requests
|
||||
from tqdm import tqdm
|
||||
import json
|
||||
from PIL import Image
|
||||
import random
|
||||
import time
|
||||
import pathlib
|
||||
import textwrap
|
||||
from argparse import ArgumentParser
|
||||
import google.generativeai as genai
|
||||
import json
|
||||
from PIL import Image
|
||||
from IPython.display import display
|
||||
from IPython.display import Markdown
|
||||
from tqdm import tqdm
|
||||
import os
|
||||
def encode_image(image_path):
|
||||
with open(image_path, "rb") as image_file:
|
||||
return base64.b64encode(image_file.read()).decode('utf-8')
|
||||
def save_json(json_list,save_path):
|
||||
with open(save_path, 'w') as file:
|
||||
json.dump(json_list, file,indent=4)
|
||||
def _get_args():
|
||||
parser = ArgumentParser()
|
||||
parser.add_argument("--image_folder", type=str, default="./data")
|
||||
parser.add_argument("--output_path", type=str, default="./results")
|
||||
parser.add_argument("--OCRBench_file", type=str, default="./OCRBench/OCRBench.json")
|
||||
parser.add_argument("--OPENAI_API_KEY", type=str, default="")
|
||||
parser.add_argument("--API_BASE", type=str, default="")
|
||||
parser.add_argument("--model", type=str, default="gpt-4-vision-preview")
|
||||
args = parser.parse_args()
|
||||
return args
|
||||
OCRBench_score = {"Regular Text Recognition":0,"Irregular Text Recognition":0,"Artistic Text Recognition":0,"Handwriting Recognition":0,
|
||||
"Digit String Recognition":0,"Non-Semantic Text Recognition":0,"Scene Text-centric VQA":0,"Doc-oriented VQA":0,"Doc-oriented VQA":0,
|
||||
"Key Information Extraction":0,"Handwritten Mathematical Expression Recognition":0}
|
||||
AllDataset_score = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
num_all = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
|
||||
if __name__ == "__main__":
|
||||
args = _get_args()
|
||||
|
||||
if os.path.exists(os.path.join(args.output_path,f"{args.model}.json")):
|
||||
data_path = os.path.join(args.output_path,f"{args.model}.json")
|
||||
else:
|
||||
data_path = args.OCRBench_file
|
||||
|
||||
with open(data_path,"r") as f:
|
||||
data = json.load(f)
|
||||
for i in tqdm(range(len(data))):
|
||||
img_path = os.path.join(args.image_folder, data[i]['image_path'])
|
||||
question = data[i]['question']
|
||||
if data[i].get("predict", 0)!=0:
|
||||
print(f"{img_path} predict exist, continue.")
|
||||
continue
|
||||
base64_image = encode_image(img_path)
|
||||
headers = {
|
||||
"Content-Type": "application/json",
|
||||
"Authorization": f"Bearer {args.OPENAI_API_KEY}"
|
||||
}
|
||||
payload = {
|
||||
"model": args.model,
|
||||
"messages": [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{
|
||||
"type": "text",
|
||||
"text": f"{question}"
|
||||
},
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {
|
||||
"url": f"data:image/jpeg;base64,{base64_image}"
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
],
|
||||
"max_tokens": 500
|
||||
}
|
||||
try:
|
||||
response = requests.post(args.API_BASE, headers=headers, json=payload)
|
||||
print(response.json())
|
||||
answer = response.json()['choices'][0]['message']['content']
|
||||
data[i]['predict'] = answer
|
||||
save_json(data, os.path.join(args.output_path,f"{args.model}.json"))
|
||||
except:
|
||||
time.sleep(100)
|
||||
print(f"{img_path} error")
|
||||
for i in range(len(data)):
|
||||
data_type = data[i]["type"]
|
||||
dataset_name = data[i]["dataset_name"]
|
||||
answers = data[i]["answers"]
|
||||
if data[i].get('predict',0)==0:
|
||||
continue
|
||||
predict = data[i]['predict']
|
||||
data[i]['result'] = 0
|
||||
if dataset_name == "HME100k":
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
save_json(data, os.path.join(args.output_path,f"{args.model}.json"))
|
||||
for i in range(len(data)):
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
OCRBench_score[data[i]['type']] += data[i]['result']
|
||||
recognition_score = OCRBench_score['Regular Text Recognition']+OCRBench_score['Irregular Text Recognition']+OCRBench_score['Artistic Text Recognition']+OCRBench_score['Handwriting Recognition']+OCRBench_score['Digit String Recognition']+OCRBench_score['Non-Semantic Text Recognition']
|
||||
Final_score = recognition_score+OCRBench_score['Scene Text-centric VQA']+OCRBench_score['Doc-oriented VQA']+OCRBench_score['Key Information Extraction']+OCRBench_score['Handwritten Mathematical Expression Recognition']
|
||||
print("###########################OCRBench##############################")
|
||||
print(f"Text Recognition(Total 300):{recognition_score}")
|
||||
print("------------------Details of Recognition Score-------------------")
|
||||
print(f"Regular Text Recognition(Total 50): {OCRBench_score['Regular Text Recognition']}")
|
||||
print(f"Irregular Text Recognition(Total 50): {OCRBench_score['Irregular Text Recognition']}")
|
||||
print(f"Artistic Text Recognition(Total 50): {OCRBench_score['Artistic Text Recognition']}")
|
||||
print(f"Handwriting Recognition(Total 50): {OCRBench_score['Handwriting Recognition']}")
|
||||
print(f"Digit String Recognition(Total 50): {OCRBench_score['Digit String Recognition']}")
|
||||
print(f"Non-Semantic Text Recognition(Total 50): {OCRBench_score['Non-Semantic Text Recognition']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Scene Text-centric VQA(Total 200): {OCRBench_score['Scene Text-centric VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Doc-oriented VQA(Total 200): {OCRBench_score['Doc-oriented VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Key Information Extraction(Total 200): {OCRBench_score['Key Information Extraction']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Handwritten Mathematical Expression Recognition(Total 100): {OCRBench_score['Handwritten Mathematical Expression Recognition']}")
|
||||
print("----------------------Final Score-------------------------------")
|
||||
print(f"Final Score(Total 1000): {Final_score}")
|
||||
115
scripts/Genimi.py
Normal file
115
scripts/Genimi.py
Normal file
@@ -0,0 +1,115 @@
|
||||
import pathlib
|
||||
import textwrap
|
||||
from argparse import ArgumentParser
|
||||
import google.generativeai as genai
|
||||
import json
|
||||
from PIL import Image
|
||||
from IPython.display import display
|
||||
from IPython.display import Markdown
|
||||
from tqdm import tqdm
|
||||
import os
|
||||
import sys
|
||||
OCRBench_score = {"Regular Text Recognition":0,"Irregular Text Recognition":0,"Artistic Text Recognition":0,"Handwriting Recognition":0,
|
||||
"Digit String Recognition":0,"Non-Semantic Text Recognition":0,"Scene Text-centric VQA":0,"Doc-oriented VQA":0,"Doc-oriented VQA":0,
|
||||
"Key Information Extraction":0,"Handwritten Mathematical Expression Recognition":0}
|
||||
AllDataset_score = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
num_all = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
def save_json(json_list,save_path):
|
||||
with open(save_path, 'w') as file:
|
||||
json.dump(json_list, file,indent=4)
|
||||
def _get_args():
|
||||
parser = ArgumentParser()
|
||||
parser.add_argument("--image_folder", type=str, default="./data")
|
||||
parser.add_argument("--output_path", type=str, default="./results")
|
||||
parser.add_argument("--OCRBench_file", type=str, default="./OCRBench/OCRBench.json")
|
||||
parser.add_argument("--GOOGLE_API_KEY", type=str, default="")
|
||||
parser.add_argument("--model", type=str, default="gemini-pro-vision")
|
||||
args = parser.parse_args()
|
||||
return args
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
args = _get_args()
|
||||
genai.configure(api_key=args.GOOGLE_API_KEY)
|
||||
model = genai.GenerativeModel(args.model)
|
||||
|
||||
if os.path.exists(os.path.join(args.output_path,f"{args.model}.json")):
|
||||
data_path = os.path.join(args.output_path,f"{args.model}.json")
|
||||
else:
|
||||
data_path = args.OCRBench_file
|
||||
|
||||
with open(data_path, "r") as f:
|
||||
data = json.load(f)
|
||||
for i in tqdm(range(len(data))):
|
||||
img_path = os.path.join(args.image_folder, data[i]['image_path'])
|
||||
question = data[i]['question']
|
||||
if data[i].get("predict", 0)!=0:
|
||||
print(f"{img_path} predict exist, continue.")
|
||||
continue
|
||||
try:
|
||||
img = Image.open(img_path).convert("RGB")
|
||||
response = model.generate_content([question, img])
|
||||
data[i]['predict'] = response.text
|
||||
save_json(data, os.path.join(args.output_path,f"{args.model}.json"))
|
||||
except:
|
||||
print(f"{img_path}: API call failed.")
|
||||
for i in range(len(data)):
|
||||
data_type = data[i]["type"]
|
||||
dataset_name = data[i]["dataset_name"]
|
||||
answers = data[i]["answers"]
|
||||
if data[i].get('predict',0)==0:
|
||||
continue
|
||||
predict = data[i]['predict']
|
||||
data[i]['result'] = 0
|
||||
if dataset_name == "HME100k":
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
save_json(data, os.path.join(args.output_path,f"{args.model}.json"))
|
||||
for i in range(len(data)):
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
OCRBench_score[data[i]['type']] += data[i]['result']
|
||||
recognition_score = OCRBench_score['Regular Text Recognition']+OCRBench_score['Irregular Text Recognition']+OCRBench_score['Artistic Text Recognition']+OCRBench_score['Handwriting Recognition']+OCRBench_score['Digit String Recognition']+OCRBench_score['Non-Semantic Text Recognition']
|
||||
Final_score = recognition_score+OCRBench_score['Scene Text-centric VQA']+OCRBench_score['Doc-oriented VQA']+OCRBench_score['Key Information Extraction']+OCRBench_score['Handwritten Mathematical Expression Recognition']
|
||||
print("###########################OCRBench##############################")
|
||||
print(f"Text Recognition(Total 300):{recognition_score}")
|
||||
print("------------------Details of Recognition Score-------------------")
|
||||
print(f"Regular Text Recognition(Total 50): {OCRBench_score['Regular Text Recognition']}")
|
||||
print(f"Irregular Text Recognition(Total 50): {OCRBench_score['Irregular Text Recognition']}")
|
||||
print(f"Artistic Text Recognition(Total 50): {OCRBench_score['Artistic Text Recognition']}")
|
||||
print(f"Handwriting Recognition(Total 50): {OCRBench_score['Handwriting Recognition']}")
|
||||
print(f"Digit String Recognition(Total 50): {OCRBench_score['Digit String Recognition']}")
|
||||
print(f"Non-Semantic Text Recognition(Total 50): {OCRBench_score['Non-Semantic Text Recognition']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Scene Text-centric VQA(Total 200): {OCRBench_score['Scene Text-centric VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Doc-oriented VQA(Total 200): {OCRBench_score['Doc-oriented VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Key Information Extraction(Total 200): {OCRBench_score['Key Information Extraction']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Handwritten Mathematical Expression Recognition(Total 100): {OCRBench_score['Handwritten Mathematical Expression Recognition']}")
|
||||
print("----------------------Final Score-------------------------------")
|
||||
print(f"Final Score(Total 1000): {Final_score}")
|
||||
210
scripts/LLaVA1_5.py
Normal file
210
scripts/LLaVA1_5.py
Normal file
@@ -0,0 +1,210 @@
|
||||
import json
|
||||
from argparse import ArgumentParser
|
||||
import torch
|
||||
import os
|
||||
import json
|
||||
from tqdm import tqdm
|
||||
from PIL import Image
|
||||
import math
|
||||
import multiprocessing
|
||||
from multiprocessing import Pool, Queue, Manager
|
||||
|
||||
from llava.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN
|
||||
from llava.conversation import conv_templates, SeparatorStyle
|
||||
from llava.model.builder import load_pretrained_model
|
||||
from llava.utils import disable_torch_init
|
||||
from llava.mm_utils import tokenizer_image_token, process_images, get_model_name_from_path
|
||||
|
||||
# https://github.com/haotian-liu/LLaVA/blob/main/llava/eval/model_vqa_loader.py
|
||||
|
||||
def split_list(lst, n):
|
||||
length = len(lst)
|
||||
avg = length // n # 每份的大小
|
||||
result = [] # 存储分割后的子列表
|
||||
for i in range(n - 1):
|
||||
result.append(lst[i*avg:(i+1)*avg])
|
||||
result.append(lst[(n-1)*avg:])
|
||||
return result
|
||||
|
||||
def save_json(json_list,save_path):
|
||||
with open(save_path, 'w') as file:
|
||||
json.dump(json_list, file,indent=4)
|
||||
|
||||
def _get_args():
|
||||
parser = ArgumentParser()
|
||||
parser.add_argument("--image_folder", type=str, default="./data")
|
||||
parser.add_argument("--output_folder", type=str, default="./results")
|
||||
parser.add_argument("--OCRBench_file", type=str, default="./OCRBench/OCRBench.json")
|
||||
parser.add_argument("--model_path", type=str, default="liuhaotian/llava-v1.5-7b")
|
||||
parser.add_argument("--model_base", type=str, default=None)
|
||||
parser.add_argument("--save_name", type=str, default="llava1_5_7b")
|
||||
parser.add_argument("--conv_mode", type=str, default="vicuna_v1")
|
||||
parser.add_argument("--num_workers", type=int, default=8)
|
||||
parser.add_argument("--temperature", type=float, default=0.0)
|
||||
parser.add_argument("--top_p", type=float, default=None)
|
||||
parser.add_argument("--num_beams", type=int, default=1)
|
||||
args = parser.parse_args()
|
||||
return args
|
||||
OCRBench_score = {"Regular Text Recognition":0,"Irregular Text Recognition":0,"Artistic Text Recognition":0,"Handwriting Recognition":0,
|
||||
"Digit String Recognition":0,"Non-Semantic Text Recognition":0,"Scene Text-centric VQA":0,"Doc-oriented VQA":0,"Doc-oriented VQA":0,
|
||||
"Key Information Extraction":0,"Handwritten Mathematical Expression Recognition":0}
|
||||
AllDataset_score = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
num_all = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
|
||||
def eval_worker(args, data, eval_id, output_queue):
|
||||
print(f"Process {eval_id} start.")
|
||||
device = f"cuda:{eval_id}"
|
||||
disable_torch_init()
|
||||
model_path = os.path.expanduser(args.model_path)
|
||||
model_name = get_model_name_from_path(model_path)
|
||||
tokenizer, model, image_processor, context_len = load_pretrained_model( model_path = model_path, model_base = args.model_base, model_name = model_name,device = device)
|
||||
if 'plain' in model_name and 'finetune' not in model_name.lower() and 'mmtag' not in args.conv_mode:
|
||||
args.conv_mode = args.conv_mode + '_mmtag'
|
||||
print(f'It seems that this is a plain model, but it is not using a mmtag prompt, auto switching to {args.conv_mode}.')
|
||||
for i in tqdm(range(len(data))):
|
||||
img_path = os.path.join(args.image_folder, data[i]['image_path'])
|
||||
qs = data[i]['question']
|
||||
qs = qs+"\nAnswer the question using a single word or phrase."
|
||||
if model.config.mm_use_im_start_end:
|
||||
qs = DEFAULT_IM_START_TOKEN + DEFAULT_IMAGE_TOKEN + DEFAULT_IM_END_TOKEN + '\n' + qs
|
||||
else:
|
||||
qs = DEFAULT_IMAGE_TOKEN + '\n' + qs
|
||||
conv = conv_templates[args.conv_mode].copy()
|
||||
conv.append_message(conv.roles[0], qs)
|
||||
conv.append_message(conv.roles[1], None)
|
||||
prompt = conv.get_prompt()
|
||||
|
||||
image = Image.open(img_path).convert('RGB')
|
||||
image_tensor = process_images([image], image_processor, model.config)
|
||||
input_ids = tokenizer_image_token(prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors='pt').unsqueeze(0)
|
||||
if data[i].get("predict", 0)!=0:
|
||||
print(f"{img_path} predict exist, continue.")
|
||||
continue
|
||||
|
||||
stop_str = conv_templates[args.conv_mode].sep if conv_templates[args.conv_mode].sep_style != SeparatorStyle.TWO else conv_templates[args.conv_mode].sep2
|
||||
input_ids = input_ids.to(device=device, non_blocking=True)
|
||||
with torch.inference_mode():
|
||||
output_ids = model.generate(
|
||||
input_ids,
|
||||
images=image_tensor.to(dtype=torch.float16, device=device, non_blocking=True),
|
||||
do_sample=True if args.temperature > 0 else False,
|
||||
temperature=args.temperature,
|
||||
top_p=args.top_p,
|
||||
num_beams=args.num_beams,
|
||||
max_new_tokens=128,
|
||||
use_cache=True)
|
||||
|
||||
input_token_len = input_ids.shape[1]
|
||||
n_diff_input_output = (input_ids != output_ids[:, :input_token_len]).sum().item()
|
||||
if n_diff_input_output > 0:
|
||||
print(f'[Warning] {n_diff_input_output} output_ids are not the same as the input_ids')
|
||||
outputs = tokenizer.batch_decode(output_ids[:, input_token_len:], skip_special_tokens=True)[0]
|
||||
outputs = outputs.strip()
|
||||
if outputs.endswith(stop_str):
|
||||
outputs = outputs[:-len(stop_str)]
|
||||
outputs = outputs.strip()
|
||||
|
||||
data[i]['predict'] = outputs
|
||||
output_queue.put({eval_id: data})
|
||||
print(f"Process {eval_id} has completed.")
|
||||
|
||||
if __name__=="__main__":
|
||||
multiprocessing.set_start_method('spawn')
|
||||
args = _get_args()
|
||||
|
||||
if os.path.exists(os.path.join(args.output_folder,f"{args.save_name}.json")):
|
||||
data_path = os.path.join(args.output_folder,f"{args.save_name}.json")
|
||||
print(f"output_path:{data_path} exist! Only generate the results that were not generated in {data_path}.")
|
||||
else:
|
||||
data_path = args.OCRBench_file
|
||||
|
||||
with open(data_path, "r") as f:
|
||||
data = json.load(f)
|
||||
|
||||
data_list = split_list(data, args.num_workers)
|
||||
output_queue = Manager().Queue()
|
||||
|
||||
pool = Pool(processes=args.num_workers)
|
||||
for i in range(len(data_list)):
|
||||
pool.apply_async(eval_worker, args=(args, data_list[i], i, output_queue))
|
||||
pool.close()
|
||||
pool.join()
|
||||
|
||||
results = {}
|
||||
while not output_queue.empty():
|
||||
result = output_queue.get()
|
||||
results.update(result)
|
||||
data = []
|
||||
for i in range(len(data_list)):
|
||||
data.extend(results[i])
|
||||
|
||||
|
||||
for i in range(len(data)):
|
||||
data_type = data[i]["type"]
|
||||
dataset_name = data[i]["dataset_name"]
|
||||
answers = data[i]["answers"]
|
||||
if data[i].get('predict',0)==0:
|
||||
continue
|
||||
predict = data[i]['predict']
|
||||
data[i]['result'] = 0
|
||||
if dataset_name == "HME100k":
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
save_json(data, os.path.join(args.output_folder,f"{args.save_name}.json"))
|
||||
if len(data)==1000:
|
||||
for i in range(len(data)):
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
OCRBench_score[data[i]['type']] += data[i]['result']
|
||||
recognition_score = OCRBench_score['Regular Text Recognition']+OCRBench_score['Irregular Text Recognition']+OCRBench_score['Artistic Text Recognition']+OCRBench_score['Handwriting Recognition']+OCRBench_score['Digit String Recognition']+OCRBench_score['Non-Semantic Text Recognition']
|
||||
Final_score = recognition_score+OCRBench_score['Scene Text-centric VQA']+OCRBench_score['Doc-oriented VQA']+OCRBench_score['Key Information Extraction']+OCRBench_score['Handwritten Mathematical Expression Recognition']
|
||||
print("###########################OCRBench##############################")
|
||||
print(f"Text Recognition(Total 300):{recognition_score}")
|
||||
print("------------------Details of Recognition Score-------------------")
|
||||
print(f"Regular Text Recognition(Total 50): {OCRBench_score['Regular Text Recognition']}")
|
||||
print(f"Irregular Text Recognition(Total 50): {OCRBench_score['Irregular Text Recognition']}")
|
||||
print(f"Artistic Text Recognition(Total 50): {OCRBench_score['Artistic Text Recognition']}")
|
||||
print(f"Handwriting Recognition(Total 50): {OCRBench_score['Handwriting Recognition']}")
|
||||
print(f"Digit String Recognition(Total 50): {OCRBench_score['Digit String Recognition']}")
|
||||
print(f"Non-Semantic Text Recognition(Total 50): {OCRBench_score['Non-Semantic Text Recognition']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Scene Text-centric VQA(Total 200): {OCRBench_score['Scene Text-centric VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Doc-oriented VQA(Total 200): {OCRBench_score['Doc-oriented VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Key Information Extraction(Total 200): {OCRBench_score['Key Information Extraction']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Handwritten Mathematical Expression Recognition(Total 100): {OCRBench_score['Handwritten Mathematical Expression Recognition']}")
|
||||
print("----------------------Final Score-------------------------------")
|
||||
print(f"Final Score(Total 1000): {Final_score}")
|
||||
else:
|
||||
for i in range(len(data)):
|
||||
num_all[data[i]['dataset_name']] += 1
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
AllDataset_score[data[i]['dataset_name']] += data[i]['result']
|
||||
for key in AllDataset_score.keys():
|
||||
print(f"{key}: {AllDataset_score[key]/float(num_all[key])}")
|
||||
163
scripts/blip2.py
Normal file
163
scripts/blip2.py
Normal file
@@ -0,0 +1,163 @@
|
||||
import json
|
||||
from argparse import ArgumentParser
|
||||
import torch
|
||||
import os
|
||||
import json
|
||||
from tqdm import tqdm
|
||||
from PIL import Image
|
||||
import math
|
||||
import multiprocessing
|
||||
from multiprocessing import Pool, Queue, Manager
|
||||
|
||||
from transformers import Blip2Processor, Blip2ForConditionalGeneration
|
||||
import torch
|
||||
# https://huggingface.co/Salesforce/blip2-opt-6.7b
|
||||
|
||||
def split_list(lst, n):
|
||||
length = len(lst)
|
||||
avg = length // n # 每份的大小
|
||||
result = [] # 存储分割后的子列表
|
||||
for i in range(n - 1):
|
||||
result.append(lst[i*avg:(i+1)*avg])
|
||||
result.append(lst[(n-1)*avg:])
|
||||
return result
|
||||
|
||||
def save_json(json_list,save_path):
|
||||
with open(save_path, 'w') as file:
|
||||
json.dump(json_list, file,indent=4)
|
||||
|
||||
def _get_args():
|
||||
parser = ArgumentParser()
|
||||
parser.add_argument("--image_folder", type=str, default="./data")
|
||||
parser.add_argument("--output_folder", type=str, default="./results")
|
||||
parser.add_argument("--OCRBench_file", type=str, default="./OCRBench/OCRBench.json")
|
||||
parser.add_argument("--model_path", type=str, default="./model_weights/blip2-opt-6.7b")
|
||||
parser.add_argument("--save_name", type=str, default="blip2_opt_6_7b")
|
||||
parser.add_argument("--num_workers", type=int, default=8)
|
||||
args = parser.parse_args()
|
||||
return args
|
||||
|
||||
OCRBench_score = {"Regular Text Recognition":0,"Irregular Text Recognition":0,"Artistic Text Recognition":0,"Handwriting Recognition":0,
|
||||
"Digit String Recognition":0,"Non-Semantic Text Recognition":0,"Scene Text-centric VQA":0,"Doc-oriented VQA":0,"Doc-oriented VQA":0,
|
||||
"Key Information Extraction":0,"Handwritten Mathematical Expression Recognition":0}
|
||||
AllDataset_score = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
num_all = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
|
||||
def eval_worker(args, data, eval_id, output_queue):
|
||||
print(f"Process {eval_id} start.")
|
||||
processor = Blip2Processor.from_pretrained(args.model_path)
|
||||
model = Blip2ForConditionalGeneration.from_pretrained(args.model_path, load_in_8bit=False, device_map={"": eval_id}, torch_dtype=torch.float16)
|
||||
for i in tqdm(range(len(data))):
|
||||
img_path = os.path.join(args.image_folder, data[i]['image_path'])
|
||||
qs = data[i]['question']
|
||||
if data[i].get("predict", 0)!=0:
|
||||
print(f"{img_path} predict exist, continue.")
|
||||
continue
|
||||
image = Image.open(img_path).convert("RGB")
|
||||
prompt = f"Question: {qs} Answer:"
|
||||
inputs = processor(images=image, text=prompt, return_tensors="pt").to(device=f"cuda:{eval_id}", dtype=torch.float16)
|
||||
generated_ids = model.generate(**inputs)
|
||||
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
|
||||
data[i]['predict'] = generated_text
|
||||
output_queue.put({eval_id: data})
|
||||
print(f"Process {eval_id} has completed.")
|
||||
|
||||
if __name__=="__main__":
|
||||
multiprocessing.set_start_method('spawn')
|
||||
args = _get_args()
|
||||
|
||||
if os.path.exists(os.path.join(args.output_folder,f"{args.save_name}.json")):
|
||||
data_path = os.path.join(args.output_folder,f"{args.save_name}.json")
|
||||
print(f"output_path:{data_path} exist! Only generate the results that were not generated in {data_path}.")
|
||||
else:
|
||||
data_path = args.OCRBench_file
|
||||
|
||||
with open(data_path, "r") as f:
|
||||
data = json.load(f)
|
||||
|
||||
data_list = split_list(data, args.num_workers)
|
||||
output_queue = Manager().Queue()
|
||||
|
||||
pool = Pool(processes=args.num_workers)
|
||||
for i in range(len(data_list)):
|
||||
pool.apply_async(eval_worker, args=(args, data_list[i], i, output_queue))
|
||||
pool.close()
|
||||
pool.join()
|
||||
|
||||
results = {}
|
||||
while not output_queue.empty():
|
||||
result = output_queue.get()
|
||||
results.update(result)
|
||||
data = []
|
||||
for i in range(len(data_list)):
|
||||
data.extend(results[i])
|
||||
|
||||
for i in range(len(data)):
|
||||
data_type = data[i]["type"]
|
||||
dataset_name = data[i]["dataset_name"]
|
||||
answers = data[i]["answers"]
|
||||
if data[i].get('predict',0)==0:
|
||||
continue
|
||||
predict = data[i]['predict']
|
||||
data[i]['result'] = 0
|
||||
if dataset_name == "HME100k":
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
save_json(data, os.path.join(args.output_folder,f"{args.save_name}.json"))
|
||||
if len(data)==1000:
|
||||
for i in range(len(data)):
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
OCRBench_score[data[i]['type']] += data[i]['result']
|
||||
recognition_score = OCRBench_score['Regular Text Recognition']+OCRBench_score['Irregular Text Recognition']+OCRBench_score['Artistic Text Recognition']+OCRBench_score['Handwriting Recognition']+OCRBench_score['Digit String Recognition']+OCRBench_score['Non-Semantic Text Recognition']
|
||||
Final_score = recognition_score+OCRBench_score['Scene Text-centric VQA']+OCRBench_score['Doc-oriented VQA']+OCRBench_score['Key Information Extraction']+OCRBench_score['Handwritten Mathematical Expression Recognition']
|
||||
print("###########################OCRBench##############################")
|
||||
print(f"Text Recognition(Total 300):{recognition_score}")
|
||||
print("------------------Details of Recognition Score-------------------")
|
||||
print(f"Regular Text Recognition(Total 50): {OCRBench_score['Regular Text Recognition']}")
|
||||
print(f"Irregular Text Recognition(Total 50): {OCRBench_score['Irregular Text Recognition']}")
|
||||
print(f"Artistic Text Recognition(Total 50): {OCRBench_score['Artistic Text Recognition']}")
|
||||
print(f"Handwriting Recognition(Total 50): {OCRBench_score['Handwriting Recognition']}")
|
||||
print(f"Digit String Recognition(Total 50): {OCRBench_score['Digit String Recognition']}")
|
||||
print(f"Non-Semantic Text Recognition(Total 50): {OCRBench_score['Non-Semantic Text Recognition']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Scene Text-centric VQA(Total 200): {OCRBench_score['Scene Text-centric VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Doc-oriented VQA(Total 200): {OCRBench_score['Doc-oriented VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Key Information Extraction(Total 200): {OCRBench_score['Key Information Extraction']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Handwritten Mathematical Expression Recognition(Total 100): {OCRBench_score['Handwritten Mathematical Expression Recognition']}")
|
||||
print("----------------------Final Score-------------------------------")
|
||||
print(f"Final Score(Total 1000): {Final_score}")
|
||||
else:
|
||||
for i in range(len(data)):
|
||||
num_all[data[i]['dataset_name']] += 1
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
AllDataset_score[data[i]['dataset_name']] += data[i]['result']
|
||||
for key in AllDataset_score.keys():
|
||||
print(f"{key}: {AllDataset_score[key]/float(num_all[key])}")
|
||||
175
scripts/blip2_vicuna_instruct.py
Normal file
175
scripts/blip2_vicuna_instruct.py
Normal file
@@ -0,0 +1,175 @@
|
||||
import json
|
||||
from argparse import ArgumentParser
|
||||
import torch
|
||||
import os
|
||||
import json
|
||||
from tqdm import tqdm
|
||||
from PIL import Image
|
||||
import math
|
||||
import multiprocessing
|
||||
from multiprocessing import Pool, Queue, Manager
|
||||
|
||||
from transformers import InstructBlipProcessor, InstructBlipForConditionalGeneration
|
||||
|
||||
# https://huggingface.co/Salesforce/instructblip-vicuna-7b
|
||||
def split_list(lst, n):
|
||||
length = len(lst)
|
||||
avg = length // n # 每份的大小
|
||||
result = [] # 存储分割后的子列表
|
||||
for i in range(n - 1):
|
||||
result.append(lst[i*avg:(i+1)*avg])
|
||||
result.append(lst[(n-1)*avg:])
|
||||
return result
|
||||
|
||||
def save_json(json_list,save_path):
|
||||
with open(save_path, 'w') as file:
|
||||
json.dump(json_list, file,indent=4)
|
||||
|
||||
def _get_args():
|
||||
parser = ArgumentParser()
|
||||
parser.add_argument("--image_folder", type=str, default="./data")
|
||||
parser.add_argument("--output_folder", type=str, default="./results")
|
||||
parser.add_argument("--OCRBench_file", type=str, default="./OCRBench/OCRBench.json")
|
||||
parser.add_argument("--model_path", type=str, default="./model_weights/instructblip-vicuna-7b")
|
||||
parser.add_argument("--save_name", type=str, default="instructblip_vicuna_7b")
|
||||
parser.add_argument("--num_workers", type=int, default=8)
|
||||
parser.add_argument("--temperature", type=float, default=0.0)
|
||||
args = parser.parse_args()
|
||||
return args
|
||||
|
||||
OCRBench_score = {"Regular Text Recognition":0,"Irregular Text Recognition":0,"Artistic Text Recognition":0,"Handwriting Recognition":0,
|
||||
"Digit String Recognition":0,"Non-Semantic Text Recognition":0,"Scene Text-centric VQA":0,"Doc-oriented VQA":0,"Doc-oriented VQA":0,
|
||||
"Key Information Extraction":0,"Handwritten Mathematical Expression Recognition":0}
|
||||
AllDataset_score = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
num_all = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
|
||||
def eval_worker(args, data, eval_id, output_queue):
|
||||
print(f"Process {eval_id} start.")
|
||||
device = f"cuda:{eval_id}"
|
||||
model = InstructBlipForConditionalGeneration.from_pretrained(args.model_path)
|
||||
processor = InstructBlipProcessor.from_pretrained(args.model_path)
|
||||
model.to(device)
|
||||
for i in tqdm(range(len(data))):
|
||||
img_path = os.path.join(args.image_folder, data[i]['image_path'])
|
||||
qs = data[i]['question']
|
||||
if data[i].get("predict", 0)!=0:
|
||||
print(f"{img_path} predict exist, continue.")
|
||||
continue
|
||||
image = Image.open(img_path).convert('RGB')
|
||||
inputs = processor(images=image, text=qs, return_tensors="pt").to(device)
|
||||
outputs = model.generate(
|
||||
**inputs,
|
||||
do_sample=False,
|
||||
num_beams=5,
|
||||
max_length=100,
|
||||
min_length=1,
|
||||
top_p=0.9,
|
||||
repetition_penalty=1.5,
|
||||
length_penalty=1.0,
|
||||
temperature=0,
|
||||
)
|
||||
generated_text = processor.batch_decode(outputs, skip_special_tokens=True)[0].strip()
|
||||
data[i]['predict'] = generated_text
|
||||
output_queue.put({eval_id: data})
|
||||
print(f"Process {eval_id} has completed.")
|
||||
|
||||
if __name__=="__main__":
|
||||
multiprocessing.set_start_method('spawn')
|
||||
args = _get_args()
|
||||
|
||||
if os.path.exists(os.path.join(args.output_folder,f"{args.save_name}.json")):
|
||||
data_path = os.path.join(args.output_folder,f"{args.save_name}.json")
|
||||
print(f"output_path:{data_path} exist! Only generate the results that were not generated in {data_path}.")
|
||||
else:
|
||||
data_path = args.OCRBench_file
|
||||
|
||||
with open(data_path, "r") as f:
|
||||
data = json.load(f)
|
||||
|
||||
data_list = split_list(data, args.num_workers)
|
||||
output_queue = Manager().Queue()
|
||||
|
||||
pool = Pool(processes=args.num_workers)
|
||||
for i in range(len(data_list)):
|
||||
pool.apply_async(eval_worker, args=(args, data_list[i], i, output_queue))
|
||||
pool.close()
|
||||
pool.join()
|
||||
|
||||
|
||||
results = {}
|
||||
while not output_queue.empty():
|
||||
result = output_queue.get()
|
||||
results.update(result)
|
||||
data = []
|
||||
for i in range(len(data_list)):
|
||||
data.extend(results[i])
|
||||
|
||||
for i in range(len(data)):
|
||||
data_type = data[i]["type"]
|
||||
dataset_name = data[i]["dataset_name"]
|
||||
answers = data[i]["answers"]
|
||||
if data[i].get('predict',0)==0:
|
||||
continue
|
||||
predict = data[i]['predict']
|
||||
data[i]['result'] = 0
|
||||
if dataset_name == "HME100k":
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
save_json(data, os.path.join(args.output_folder,f"{args.save_name}.json"))
|
||||
if len(data)==1000:
|
||||
for i in range(len(data)):
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
OCRBench_score[data[i]['type']] += data[i]['result']
|
||||
recognition_score = OCRBench_score['Regular Text Recognition']+OCRBench_score['Irregular Text Recognition']+OCRBench_score['Artistic Text Recognition']+OCRBench_score['Handwriting Recognition']+OCRBench_score['Digit String Recognition']+OCRBench_score['Non-Semantic Text Recognition']
|
||||
Final_score = recognition_score+OCRBench_score['Scene Text-centric VQA']+OCRBench_score['Doc-oriented VQA']+OCRBench_score['Key Information Extraction']+OCRBench_score['Handwritten Mathematical Expression Recognition']
|
||||
print("###########################OCRBench##############################")
|
||||
print(f"Text Recognition(Total 300):{recognition_score}")
|
||||
print("------------------Details of Recognition Score-------------------")
|
||||
print(f"Regular Text Recognition(Total 50): {OCRBench_score['Regular Text Recognition']}")
|
||||
print(f"Irregular Text Recognition(Total 50): {OCRBench_score['Irregular Text Recognition']}")
|
||||
print(f"Artistic Text Recognition(Total 50): {OCRBench_score['Artistic Text Recognition']}")
|
||||
print(f"Handwriting Recognition(Total 50): {OCRBench_score['Handwriting Recognition']}")
|
||||
print(f"Digit String Recognition(Total 50): {OCRBench_score['Digit String Recognition']}")
|
||||
print(f"Non-Semantic Text Recognition(Total 50): {OCRBench_score['Non-Semantic Text Recognition']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Scene Text-centric VQA(Total 200): {OCRBench_score['Scene Text-centric VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Doc-oriented VQA(Total 200): {OCRBench_score['Doc-oriented VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Key Information Extraction(Total 200): {OCRBench_score['Key Information Extraction']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Handwritten Mathematical Expression Recognition(Total 100): {OCRBench_score['Handwritten Mathematical Expression Recognition']}")
|
||||
print("----------------------Final Score-------------------------------")
|
||||
print(f"Final Score(Total 1000): {Final_score}")
|
||||
else:
|
||||
for i in range(len(data)):
|
||||
num_all[data[i]['dataset_name']] += 1
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
AllDataset_score[data[i]['dataset_name']] += data[i]['result']
|
||||
for key in AllDataset_score.keys():
|
||||
print(f"{key}: {AllDataset_score[key]/float(num_all[key])}")
|
||||
179
scripts/bliva.py
Normal file
179
scripts/bliva.py
Normal file
@@ -0,0 +1,179 @@
|
||||
import json
|
||||
from argparse import ArgumentParser
|
||||
import torch
|
||||
import os
|
||||
import json
|
||||
from tqdm import tqdm
|
||||
from PIL import Image
|
||||
import math
|
||||
import multiprocessing
|
||||
from multiprocessing import Pool, Queue, Manager
|
||||
from bliva.models import load_model_and_preprocess
|
||||
import numpy as np
|
||||
|
||||
# https://github.com/mlpc-ucsd/BLIVA/blob/main/evaluate.py
|
||||
|
||||
def disable_torch_init():
|
||||
"""
|
||||
Disable the redundant torch default initialization to accelerate model creation.
|
||||
"""
|
||||
import torch
|
||||
setattr(torch.nn.Linear, "reset_parameters", lambda self: None)
|
||||
setattr(torch.nn.LayerNorm, "reset_parameters", lambda self: None)
|
||||
|
||||
def split_list(lst, n):
|
||||
length = len(lst)
|
||||
avg = length // n # 每份的大小
|
||||
result = [] # 存储分割后的子列表
|
||||
for i in range(n - 1):
|
||||
result.append(lst[i*avg:(i+1)*avg])
|
||||
result.append(lst[(n-1)*avg:])
|
||||
return result
|
||||
|
||||
def save_json(json_list,save_path):
|
||||
with open(save_path, 'w') as file:
|
||||
json.dump(json_list, file,indent=4)
|
||||
|
||||
def _get_args():
|
||||
parser = ArgumentParser()
|
||||
parser.add_argument("--image_folder", type=str, default="./data")
|
||||
parser.add_argument("--output_folder", type=str, default="./results")
|
||||
parser.add_argument("--OCRBench_file", type=str, default="./OCRBench/OCRBench.json")
|
||||
parser.add_argument("--model_path", type=str, default="bliva_vicuna")
|
||||
parser.add_argument("--save_name", type=str, default="bliva")
|
||||
parser.add_argument("--num_workers", type=int, default=8)
|
||||
args = parser.parse_args()
|
||||
return args
|
||||
|
||||
OCRBench_score = {"Regular Text Recognition":0,"Irregular Text Recognition":0,"Artistic Text Recognition":0,"Handwriting Recognition":0,
|
||||
"Digit String Recognition":0,"Non-Semantic Text Recognition":0,"Scene Text-centric VQA":0,"Doc-oriented VQA":0,"Doc-oriented VQA":0,
|
||||
"Key Information Extraction":0,"Handwritten Mathematical Expression Recognition":0}
|
||||
AllDataset_score = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
num_all = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
|
||||
def eval_worker(args, data, eval_id, output_queue):
|
||||
print(f"Process {eval_id} start.")
|
||||
device = f"cuda:{eval_id}"
|
||||
np.random.seed(0)
|
||||
disable_torch_init()
|
||||
if "vicuna" in args.model_path.lower():
|
||||
print("load bliva-vicuna")
|
||||
model, vis_processors, _ = load_model_and_preprocess(name=args.model_path, model_type="vicuna7b", is_eval=True, device=device)
|
||||
if "flant5xxl" in args.model_path.lower():
|
||||
print("load bliva-flant5xxl")
|
||||
model, vis_processors, _ = load_model_and_preprocess(name=args.model_path, model_type="flant5xxl", is_eval=True, device=device)
|
||||
vis_processor = vis_processors["eval"]
|
||||
for i in tqdm(range(len(data))):
|
||||
img_path = os.path.join(args.image_folder, data[i]['image_path'])
|
||||
qs = data[i]['question']
|
||||
if data[i].get("predict", 0)!=0:
|
||||
print(f"{img_path} predict exist, continue.")
|
||||
continue
|
||||
image = Image.open(img_path).convert('RGB')
|
||||
question = [qs]
|
||||
image = vis_processor(image).unsqueeze(0).to(device)
|
||||
outputs = model.generate({"image": image, "prompt": qs}, max_length=150)
|
||||
data[i]['predict'] = outputs[0].split('### Assistant:')[0]
|
||||
output_queue.put({eval_id: data})
|
||||
print(f"Process {eval_id} has completed.")
|
||||
|
||||
if __name__=="__main__":
|
||||
multiprocessing.set_start_method('spawn')
|
||||
args = _get_args()
|
||||
|
||||
if os.path.exists(os.path.join(args.output_folder,f"{args.save_name}.json")):
|
||||
data_path = os.path.join(args.output_folder,f"{args.save_name}.json")
|
||||
print(f"output_path:{data_path} exist! Only generate the results that were not generated in {data_path}.")
|
||||
else:
|
||||
data_path = args.OCRBench_file
|
||||
|
||||
with open(data_path, "r") as f:
|
||||
data = json.load(f)
|
||||
|
||||
data_list = split_list(data, args.num_workers)
|
||||
output_queue = Manager().Queue()
|
||||
|
||||
pool = Pool(processes=args.num_workers)
|
||||
for i in range(len(data_list)):
|
||||
pool.apply_async(eval_worker, args=(args, data_list[i], i, output_queue))
|
||||
pool.close()
|
||||
pool.join()
|
||||
|
||||
|
||||
results = {}
|
||||
while not output_queue.empty():
|
||||
result = output_queue.get()
|
||||
results.update(result)
|
||||
data = []
|
||||
for i in range(len(data_list)):
|
||||
data.extend(results[i])
|
||||
|
||||
for i in range(len(data)):
|
||||
data_type = data[i]["type"]
|
||||
dataset_name = data[i]["dataset_name"]
|
||||
answers = data[i]["answers"]
|
||||
if data[i].get('predict',0)==0:
|
||||
continue
|
||||
predict = data[i]['predict']
|
||||
data[i]['result'] = 0
|
||||
if dataset_name == "HME100k":
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
save_json(data, os.path.join(args.output_folder,f"{args.save_name}.json"))
|
||||
if len(data)==1000:
|
||||
for i in range(len(data)):
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
OCRBench_score[data[i]['type']] += data[i]['result']
|
||||
recognition_score = OCRBench_score['Regular Text Recognition']+OCRBench_score['Irregular Text Recognition']+OCRBench_score['Artistic Text Recognition']+OCRBench_score['Handwriting Recognition']+OCRBench_score['Digit String Recognition']+OCRBench_score['Non-Semantic Text Recognition']
|
||||
Final_score = recognition_score+OCRBench_score['Scene Text-centric VQA']+OCRBench_score['Doc-oriented VQA']+OCRBench_score['Key Information Extraction']+OCRBench_score['Handwritten Mathematical Expression Recognition']
|
||||
print("###########################OCRBench##############################")
|
||||
print(f"Text Recognition(Total 300):{recognition_score}")
|
||||
print("------------------Details of Recognition Score-------------------")
|
||||
print(f"Regular Text Recognition(Total 50): {OCRBench_score['Regular Text Recognition']}")
|
||||
print(f"Irregular Text Recognition(Total 50): {OCRBench_score['Irregular Text Recognition']}")
|
||||
print(f"Artistic Text Recognition(Total 50): {OCRBench_score['Artistic Text Recognition']}")
|
||||
print(f"Handwriting Recognition(Total 50): {OCRBench_score['Handwriting Recognition']}")
|
||||
print(f"Digit String Recognition(Total 50): {OCRBench_score['Digit String Recognition']}")
|
||||
print(f"Non-Semantic Text Recognition(Total 50): {OCRBench_score['Non-Semantic Text Recognition']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Scene Text-centric VQA(Total 200): {OCRBench_score['Scene Text-centric VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Doc-oriented VQA(Total 200): {OCRBench_score['Doc-oriented VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Key Information Extraction(Total 200): {OCRBench_score['Key Information Extraction']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Handwritten Mathematical Expression Recognition(Total 100): {OCRBench_score['Handwritten Mathematical Expression Recognition']}")
|
||||
print("----------------------Final Score-------------------------------")
|
||||
print(f"Final Score(Total 1000): {Final_score}")
|
||||
else:
|
||||
for i in range(len(data)):
|
||||
num_all[data[i]['dataset_name']] += 1
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
AllDataset_score[data[i]['dataset_name']] += data[i]['result']
|
||||
for key in AllDataset_score.keys():
|
||||
print(f"{key}: {AllDataset_score[key]/float(num_all[key])}")
|
||||
330
scripts/llavar.py
Normal file
330
scripts/llavar.py
Normal file
@@ -0,0 +1,330 @@
|
||||
import json
|
||||
from argparse import ArgumentParser
|
||||
import torch
|
||||
import os
|
||||
import json
|
||||
from tqdm import tqdm
|
||||
from PIL import Image
|
||||
import math
|
||||
import multiprocessing
|
||||
from multiprocessing import Pool, Queue, Manager
|
||||
|
||||
from transformers import AutoTokenizer, AutoModelForCausalLM, AutoConfig
|
||||
from llava import LlavaLlamaForCausalLM
|
||||
from llava.conversation import conv_templates
|
||||
from llava import conversation as conversation_lib
|
||||
from llava.utils import disable_torch_init
|
||||
from transformers import CLIPVisionModel, CLIPImageProcessor, StoppingCriteria
|
||||
from PIL import Image,ImageOps
|
||||
# https://github.com/SALT-NLP/LLaVAR/blob/main/LLaVA/llava/eval/model_vqa.py
|
||||
|
||||
def resize_image(image, target_size):
|
||||
width, height = image.size
|
||||
aspect_ratio = width / height
|
||||
if aspect_ratio > 1:
|
||||
new_width = target_size[0]
|
||||
new_height = int(new_width / aspect_ratio)
|
||||
else:
|
||||
new_height = target_size[1]
|
||||
new_width = int(new_height * aspect_ratio)
|
||||
image = image.resize((new_width, new_height))
|
||||
width_diff = target_size[0] - image.size[0]
|
||||
height_diff = target_size[1] - image.size[1]
|
||||
left_padding = 0
|
||||
top_padding = 0
|
||||
right_padding = width_diff - left_padding
|
||||
bottom_padding = height_diff - top_padding
|
||||
padded_image = ImageOps.expand(image, border=(left_padding, top_padding, right_padding, bottom_padding), fill=0)
|
||||
return padded_image
|
||||
|
||||
DEFAULT_IMAGE_TOKEN = "<image>"
|
||||
DEFAULT_IMAGE_PATCH_TOKEN = "<im_patch>"
|
||||
DEFAULT_IM_START_TOKEN = "<im_start>"
|
||||
DEFAULT_IM_END_TOKEN = "<im_end>"
|
||||
|
||||
def patch_config(config):
|
||||
patch_dict = {
|
||||
"use_mm_proj": True,
|
||||
"mm_vision_tower": "openai/clip-vit-large-patch14",
|
||||
"mm_hidden_size": 1024
|
||||
}
|
||||
|
||||
cfg = AutoConfig.from_pretrained(config)
|
||||
if not hasattr(cfg, "mm_vision_tower"):
|
||||
print(f'`mm_vision_tower` not found in `{config}`, applying patch and save to disk.')
|
||||
for k, v in patch_dict.items():
|
||||
setattr(cfg, k, v)
|
||||
cfg.save_pretrained(config)
|
||||
|
||||
def split_list(lst, n):
|
||||
length = len(lst)
|
||||
avg = length // n # 每份的大小
|
||||
result = [] # 存储分割后的子列表
|
||||
for i in range(n - 1):
|
||||
result.append(lst[i*avg:(i+1)*avg])
|
||||
result.append(lst[(n-1)*avg:])
|
||||
return result
|
||||
|
||||
def save_json(json_list,save_path):
|
||||
with open(save_path, 'w') as file:
|
||||
json.dump(json_list, file,indent=4)
|
||||
|
||||
def _get_args():
|
||||
parser = ArgumentParser()
|
||||
parser.add_argument("--image_folder", type=str, default="./data")
|
||||
parser.add_argument("--output_folder", type=str, default="./results")
|
||||
parser.add_argument("--OCRBench_file", type=str, default="./OCRBench/OCRBench.json")
|
||||
parser.add_argument("--model_path", type=str, default="./model_weights/LLaVar")
|
||||
parser.add_argument("--save_name", type=str, default="llavar")
|
||||
parser.add_argument("--conv-mode", type=str, default="llava_v1")
|
||||
parser.add_argument("--mm-projector", type=str, default=None)
|
||||
parser.add_argument("--vision-tower", type=str, default=None)
|
||||
parser.add_argument("--num_workers", type=int, default=8)
|
||||
args = parser.parse_args()
|
||||
return args
|
||||
OCRBench_score = {"Regular Text Recognition":0,"Irregular Text Recognition":0,"Artistic Text Recognition":0,"Handwriting Recognition":0,
|
||||
"Digit String Recognition":0,"Non-Semantic Text Recognition":0,"Scene Text-centric VQA":0,"Doc-oriented VQA":0,"Doc-oriented VQA":0,
|
||||
"Key Information Extraction":0,"Handwritten Mathematical Expression Recognition":0}
|
||||
AllDataset_score = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
num_all = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
|
||||
def eval_worker(args, data, eval_id, output_queue):
|
||||
print(f"Process {eval_id} start.")
|
||||
device = f"cuda:{eval_id}"
|
||||
disable_torch_init()
|
||||
model_name = os.path.expanduser(args.model_path)
|
||||
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
||||
if args.mm_projector is None:
|
||||
patch_config(model_name)
|
||||
model = LlavaLlamaForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16).to(device)
|
||||
image_processor = CLIPImageProcessor.from_pretrained(model.config.mm_vision_tower, torch_dtype=torch.float16)
|
||||
|
||||
mm_use_im_start_end = getattr(model.config, "mm_use_im_start_end", False)
|
||||
tokenizer.add_tokens([DEFAULT_IMAGE_PATCH_TOKEN], special_tokens=True)
|
||||
if mm_use_im_start_end:
|
||||
tokenizer.add_tokens([DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN], special_tokens=True)
|
||||
|
||||
vision_tower = model.model.vision_tower[0]
|
||||
vision_tower.to(device=device, dtype=torch.float16)
|
||||
vision_config = vision_tower.config
|
||||
vision_config.im_patch_token = tokenizer.convert_tokens_to_ids([DEFAULT_IMAGE_PATCH_TOKEN])[0]
|
||||
vision_config.use_im_start_end = mm_use_im_start_end
|
||||
if mm_use_im_start_end:
|
||||
vision_config.im_start_token, vision_config.im_end_token = tokenizer.convert_tokens_to_ids([DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN])
|
||||
image_token_len = (vision_config.image_size // vision_config.patch_size) ** 2
|
||||
else:
|
||||
# in case of using a pretrained model with only a MLP projector weights
|
||||
model = LlavaLlamaForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16).to(device)
|
||||
|
||||
vision_tower = CLIPVisionModel.from_pretrained(args.vision_tower, torch_dtype=torch.float16).to(device)
|
||||
image_processor = CLIPImageProcessor.from_pretrained(args.vision_tower, torch_dtype=torch.float16)
|
||||
|
||||
mm_use_im_start_end = getattr(model.config, "mm_use_im_start_end", False)
|
||||
tokenizer.add_tokens([DEFAULT_IMAGE_PATCH_TOKEN], special_tokens=True)
|
||||
if mm_use_im_start_end:
|
||||
tokenizer.add_tokens([DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN], special_tokens=True)
|
||||
|
||||
vision_config = vision_tower.config
|
||||
vision_config.im_patch_token = tokenizer.convert_tokens_to_ids([DEFAULT_IMAGE_PATCH_TOKEN])[0]
|
||||
vision_config.use_im_start_end = mm_use_im_start_end
|
||||
if mm_use_im_start_end:
|
||||
vision_config.im_start_token, vision_config.im_end_token = tokenizer.convert_tokens_to_ids([DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN])
|
||||
|
||||
image_token_len = (vision_config.image_size // vision_config.patch_size) ** 2
|
||||
|
||||
mm_projector = torch.nn.Linear(vision_config.hidden_size, model.config.hidden_size)
|
||||
mm_projector_weights = torch.load(args.mm_projector, map_location='cpu')
|
||||
mm_projector.load_state_dict({k.split('.')[-1]: v for k, v in mm_projector_weights.items()})
|
||||
|
||||
model.model.mm_projector = mm_projector.to(device).half()
|
||||
model.model.vision_tower = [vision_tower]
|
||||
for i in tqdm(range(len(data))):
|
||||
img_path = os.path.join(args.image_folder, data[i]['image_path'])
|
||||
qs = data[i]['question']
|
||||
# qs = qs+"\nAnswer the question using a single word or phrase."
|
||||
if data[i].get("predict", 0)!=0:
|
||||
print(f"{img_path} predict exist, continue.")
|
||||
continue
|
||||
if mm_use_im_start_end:
|
||||
qs = qs + '\n' + DEFAULT_IM_START_TOKEN + DEFAULT_IMAGE_PATCH_TOKEN * image_token_len + DEFAULT_IM_END_TOKEN
|
||||
else:
|
||||
qs = qs + '\n' + DEFAULT_IMAGE_PATCH_TOKEN * image_token_len
|
||||
if args.conv_mode == 'simple_legacy':
|
||||
qs += '\n\n### Response:'
|
||||
# conv = default_conversation.copy()
|
||||
conv = conv_templates[args.conv_mode].copy()
|
||||
conv.append_message(conv.roles[0], qs)
|
||||
# modified
|
||||
conv.append_message(conv.roles[1], None)
|
||||
prompt = conv.get_prompt()
|
||||
inputs = tokenizer([prompt])
|
||||
image = Image.open(img_path)
|
||||
# if "REval" in args.image_folder:
|
||||
image = resize_image(image, (336, 336))
|
||||
|
||||
image_tensor = image_processor.preprocess(image, return_tensors='pt')['pixel_values'][0]
|
||||
|
||||
input_ids = torch.as_tensor(inputs.input_ids).to(device)
|
||||
|
||||
# new stopping implementation
|
||||
class KeywordsStoppingCriteria(StoppingCriteria):
|
||||
def __init__(self, keywords, tokenizer, input_ids):
|
||||
self.keywords = keywords
|
||||
self.tokenizer = tokenizer
|
||||
self.start_len = None
|
||||
self.input_ids = input_ids
|
||||
|
||||
def __call__(self, output_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
|
||||
if self.start_len is None:
|
||||
self.start_len = self.input_ids.shape[1]
|
||||
else:
|
||||
outputs = self.tokenizer.batch_decode(output_ids[:, self.start_len:], skip_special_tokens=True)[0]
|
||||
for keyword in self.keywords:
|
||||
if keyword in outputs:
|
||||
return True
|
||||
return False
|
||||
|
||||
# keywords = ['###']
|
||||
# modified
|
||||
keywords = ['</s>']
|
||||
stopping_criteria = KeywordsStoppingCriteria(keywords, tokenizer, input_ids)
|
||||
|
||||
with torch.inference_mode():
|
||||
output_ids = model.generate(
|
||||
input_ids,
|
||||
images=image_tensor.unsqueeze(0).half().to(device),
|
||||
do_sample=False,
|
||||
temperature=0,
|
||||
max_new_tokens=200,
|
||||
stopping_criteria=[stopping_criteria])
|
||||
input_token_len = input_ids.shape[1]
|
||||
n_diff_input_output = (input_ids != output_ids[:, :input_token_len]).sum().item()
|
||||
if n_diff_input_output > 0:
|
||||
print(f'[Warning] Sample {i}: {n_diff_input_output} output_ids are not the same as the input_ids')
|
||||
outputs = tokenizer.batch_decode(output_ids[:, input_token_len:], skip_special_tokens=True)[0]
|
||||
|
||||
# modified
|
||||
if args.conv_mode == 'simple_legacy' or args.conv_mode == 'simple':
|
||||
while True:
|
||||
cur_len = len(outputs)
|
||||
outputs = outputs.strip()
|
||||
for pattern in ['###', 'Assistant:', 'Response:']:
|
||||
if outputs.startswith(pattern):
|
||||
outputs = outputs[len(pattern):].strip()
|
||||
if len(outputs) == cur_len:
|
||||
break
|
||||
|
||||
if conv.sep_style == conversation_lib.SeparatorStyle.TWO:
|
||||
sep = conv.sep2
|
||||
else:
|
||||
sep = conv.sep
|
||||
|
||||
try:
|
||||
index = outputs.index(sep)
|
||||
except ValueError:
|
||||
outputs += sep
|
||||
index = outputs.index(sep)
|
||||
outputs = outputs[:index].strip()
|
||||
data[i]['predict'] = outputs
|
||||
output_queue.put({eval_id: data})
|
||||
print(f"Process {eval_id} has completed.")
|
||||
|
||||
if __name__=="__main__":
|
||||
multiprocessing.set_start_method('spawn')
|
||||
args = _get_args()
|
||||
|
||||
if os.path.exists(os.path.join(args.output_folder,f"{args.save_name}.json")):
|
||||
data_path = os.path.join(args.output_folder,f"{args.save_name}.json")
|
||||
print(f"output_path:{data_path} exist! Only generate the results that were not generated in {data_path}.")
|
||||
else:
|
||||
data_path = args.OCRBench_file
|
||||
|
||||
with open(data_path, "r") as f:
|
||||
data = json.load(f)
|
||||
|
||||
data_list = split_list(data, args.num_workers)
|
||||
output_queue = Manager().Queue()
|
||||
|
||||
pool = Pool(processes=args.num_workers)
|
||||
for i in range(len(data_list)):
|
||||
pool.apply_async(eval_worker, args=(args, data_list[i], i, output_queue))
|
||||
pool.close()
|
||||
pool.join()
|
||||
|
||||
results = {}
|
||||
while not output_queue.empty():
|
||||
result = output_queue.get()
|
||||
results.update(result)
|
||||
data = []
|
||||
for i in range(len(data_list)):
|
||||
data.extend(results[i])
|
||||
|
||||
for i in range(len(data)):
|
||||
data_type = data[i]["type"]
|
||||
dataset_name = data[i]["dataset_name"]
|
||||
answers = data[i]["answers"]
|
||||
if data[i].get('predict',0)==0:
|
||||
continue
|
||||
predict = data[i]['predict']
|
||||
data[i]['result'] = 0
|
||||
if dataset_name == "HME100k":
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
save_json(data, os.path.join(args.output_folder,f"{args.save_name}.json"))
|
||||
if len(data)==1000:
|
||||
for i in range(len(data)):
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
OCRBench_score[data[i]['type']] += data[i]['result']
|
||||
recognition_score = OCRBench_score['Regular Text Recognition']+OCRBench_score['Irregular Text Recognition']+OCRBench_score['Artistic Text Recognition']+OCRBench_score['Handwriting Recognition']+OCRBench_score['Digit String Recognition']+OCRBench_score['Non-Semantic Text Recognition']
|
||||
Final_score = recognition_score+OCRBench_score['Scene Text-centric VQA']+OCRBench_score['Doc-oriented VQA']+OCRBench_score['Key Information Extraction']+OCRBench_score['Handwritten Mathematical Expression Recognition']
|
||||
print("###########################OCRBench##############################")
|
||||
print(f"Text Recognition(Total 300):{recognition_score}")
|
||||
print("------------------Details of Recognition Score-------------------")
|
||||
print(f"Regular Text Recognition(Total 50): {OCRBench_score['Regular Text Recognition']}")
|
||||
print(f"Irregular Text Recognition(Total 50): {OCRBench_score['Irregular Text Recognition']}")
|
||||
print(f"Artistic Text Recognition(Total 50): {OCRBench_score['Artistic Text Recognition']}")
|
||||
print(f"Handwriting Recognition(Total 50): {OCRBench_score['Handwriting Recognition']}")
|
||||
print(f"Digit String Recognition(Total 50): {OCRBench_score['Digit String Recognition']}")
|
||||
print(f"Non-Semantic Text Recognition(Total 50): {OCRBench_score['Non-Semantic Text Recognition']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Scene Text-centric VQA(Total 200): {OCRBench_score['Scene Text-centric VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Doc-oriented VQA(Total 200): {OCRBench_score['Doc-oriented VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Key Information Extraction(Total 200): {OCRBench_score['Key Information Extraction']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Handwritten Mathematical Expression Recognition(Total 100): {OCRBench_score['Handwritten Mathematical Expression Recognition']}")
|
||||
print("----------------------Final Score-------------------------------")
|
||||
print(f"Final Score(Total 1000): {Final_score}")
|
||||
else:
|
||||
for i in range(len(data)):
|
||||
num_all[data[i]['dataset_name']] += 1
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
AllDataset_score[data[i]['dataset_name']] += data[i]['result']
|
||||
for key in AllDataset_score.keys():
|
||||
print(f"{key}: {AllDataset_score[key]/float(num_all[key])}")
|
||||
185
scripts/mPLUG-owl.py
Normal file
185
scripts/mPLUG-owl.py
Normal file
@@ -0,0 +1,185 @@
|
||||
import json
|
||||
from argparse import ArgumentParser
|
||||
import torch
|
||||
import os
|
||||
import json
|
||||
from tqdm import tqdm
|
||||
from PIL import Image
|
||||
import math
|
||||
import multiprocessing
|
||||
from multiprocessing import Pool, Queue, Manager
|
||||
|
||||
import sys
|
||||
sys.path.append("./scripts/mPLUG-Owl/mPLUG-Owl/")
|
||||
from mplug_owl.modeling_mplug_owl import MplugOwlForConditionalGeneration
|
||||
from mplug_owl.tokenization_mplug_owl import MplugOwlTokenizer
|
||||
from mplug_owl.processing_mplug_owl import MplugOwlImageProcessor, MplugOwlProcessor
|
||||
|
||||
# https://github.com/X-PLUG/mPLUG-Owl/tree/main/mPLUG-Owl
|
||||
def split_list(lst, n):
|
||||
length = len(lst)
|
||||
avg = length // n # 每份的大小
|
||||
result = [] # 存储分割后的子列表
|
||||
for i in range(n - 1):
|
||||
result.append(lst[i*avg:(i+1)*avg])
|
||||
result.append(lst[(n-1)*avg:])
|
||||
return result
|
||||
|
||||
def save_json(json_list,save_path):
|
||||
with open(save_path, 'w') as file:
|
||||
json.dump(json_list, file,indent=4)
|
||||
|
||||
def _get_args():
|
||||
parser = ArgumentParser()
|
||||
parser.add_argument("--image_folder", type=str, default="./data")
|
||||
parser.add_argument("--output_folder", type=str, default="./results")
|
||||
parser.add_argument("--OCRBench_file", type=str, default="./OCRBench/OCRBench.json")
|
||||
parser.add_argument("--model_path", type=str, default="./model_weights/mplug-owl")
|
||||
parser.add_argument("--save_name", type=str, default="mplug-owl")
|
||||
parser.add_argument("--num_workers", type=int, default=8)
|
||||
args = parser.parse_args()
|
||||
return args
|
||||
|
||||
OCRBench_score = {"Regular Text Recognition":0,"Irregular Text Recognition":0,"Artistic Text Recognition":0,"Handwriting Recognition":0,
|
||||
"Digit String Recognition":0,"Non-Semantic Text Recognition":0,"Scene Text-centric VQA":0,"Doc-oriented VQA":0,"Doc-oriented VQA":0,
|
||||
"Key Information Extraction":0,"Handwritten Mathematical Expression Recognition":0}
|
||||
AllDataset_score = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
num_all = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
|
||||
def eval_worker(args, data, eval_id, output_queue):
|
||||
print(f"Process {eval_id} start.")
|
||||
pretrained_ckpt = args.model_path
|
||||
model = MplugOwlForConditionalGeneration.from_pretrained(
|
||||
pretrained_ckpt,
|
||||
torch_dtype=torch.bfloat16,
|
||||
)
|
||||
model.to(f"cuda:{eval_id}")
|
||||
image_processor = MplugOwlImageProcessor.from_pretrained(pretrained_ckpt)
|
||||
tokenizer = MplugOwlTokenizer.from_pretrained(pretrained_ckpt)
|
||||
processor = MplugOwlProcessor(image_processor, tokenizer)
|
||||
for i in tqdm(range(len(data))):
|
||||
img_path = os.path.join(args.image_folder, data[i]['image_path'])
|
||||
qs = data[i]['question']
|
||||
prompts = [
|
||||
f'''The following is a conversation between a curious human and AI assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
|
||||
Human: <image>
|
||||
Human: {qs}
|
||||
AI: ''']
|
||||
if data[i].get("predict", 0)!=0:
|
||||
print(f"{img_path} predict exist, continue.")
|
||||
continue
|
||||
generate_kwargs = {
|
||||
'do_sample': False,
|
||||
'top_k': 1,
|
||||
'max_length': 100
|
||||
}
|
||||
images = [Image.open(img_path)]
|
||||
inputs = processor(text=prompts, images=images, return_tensors='pt')
|
||||
inputs = {k: v.bfloat16() if v.dtype == torch.float else v for k, v in inputs.items()}
|
||||
inputs = {k: v.to(model.device) for k, v in inputs.items()}
|
||||
with torch.no_grad():
|
||||
res = model.generate(**inputs, **generate_kwargs)
|
||||
sentence = tokenizer.decode(res.tolist()[0], skip_special_tokens=True)
|
||||
data[i]['predict'] = sentence
|
||||
output_queue.put({eval_id: data})
|
||||
print(f"Process {eval_id} has completed.")
|
||||
|
||||
if __name__=="__main__":
|
||||
multiprocessing.set_start_method('spawn')
|
||||
args = _get_args()
|
||||
|
||||
if os.path.exists(os.path.join(args.output_folder,f"{args.save_name}.json")):
|
||||
data_path = os.path.join(args.output_folder,f"{args.save_name}.json")
|
||||
print(f"output_path:{data_path} exist! Only generate the results that were not generated in {data_path}.")
|
||||
else:
|
||||
data_path = args.OCRBench_file
|
||||
|
||||
with open(data_path, "r") as f:
|
||||
data = json.load(f)
|
||||
|
||||
data_list = split_list(data, args.num_workers)
|
||||
output_queue = Manager().Queue()
|
||||
|
||||
pool = Pool(processes=args.num_workers)
|
||||
for i in range(len(data_list)):
|
||||
pool.apply_async(eval_worker, args=(args, data_list[i], i, output_queue))
|
||||
pool.close()
|
||||
pool.join()
|
||||
|
||||
results = {}
|
||||
while not output_queue.empty():
|
||||
result = output_queue.get()
|
||||
results.update(result)
|
||||
data = []
|
||||
for i in range(len(data_list)):
|
||||
data.extend(results[i])
|
||||
|
||||
for i in range(len(data)):
|
||||
data_type = data[i]["type"]
|
||||
dataset_name = data[i]["dataset_name"]
|
||||
answers = data[i]["answers"]
|
||||
if data[i].get('predict',0)==0:
|
||||
continue
|
||||
predict = data[i]['predict']
|
||||
data[i]['result'] = 0
|
||||
if dataset_name == "HME100k":
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
save_json(data, os.path.join(args.output_folder,f"{args.save_name}.json"))
|
||||
if len(data)==1000:
|
||||
for i in range(len(data)):
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
OCRBench_score[data[i]['type']] += data[i]['result']
|
||||
recognition_score = OCRBench_score['Regular Text Recognition']+OCRBench_score['Irregular Text Recognition']+OCRBench_score['Artistic Text Recognition']+OCRBench_score['Handwriting Recognition']+OCRBench_score['Digit String Recognition']+OCRBench_score['Non-Semantic Text Recognition']
|
||||
Final_score = recognition_score+OCRBench_score['Scene Text-centric VQA']+OCRBench_score['Doc-oriented VQA']+OCRBench_score['Key Information Extraction']+OCRBench_score['Handwritten Mathematical Expression Recognition']
|
||||
print("###########################OCRBench##############################")
|
||||
print(f"Text Recognition(Total 300):{recognition_score}")
|
||||
print("------------------Details of Recognition Score-------------------")
|
||||
print(f"Regular Text Recognition(Total 50): {OCRBench_score['Regular Text Recognition']}")
|
||||
print(f"Irregular Text Recognition(Total 50): {OCRBench_score['Irregular Text Recognition']}")
|
||||
print(f"Artistic Text Recognition(Total 50): {OCRBench_score['Artistic Text Recognition']}")
|
||||
print(f"Handwriting Recognition(Total 50): {OCRBench_score['Handwriting Recognition']}")
|
||||
print(f"Digit String Recognition(Total 50): {OCRBench_score['Digit String Recognition']}")
|
||||
print(f"Non-Semantic Text Recognition(Total 50): {OCRBench_score['Non-Semantic Text Recognition']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Scene Text-centric VQA(Total 200): {OCRBench_score['Scene Text-centric VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Doc-oriented VQA(Total 200): {OCRBench_score['Doc-oriented VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Key Information Extraction(Total 200): {OCRBench_score['Key Information Extraction']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Handwritten Mathematical Expression Recognition(Total 100): {OCRBench_score['Handwritten Mathematical Expression Recognition']}")
|
||||
print("----------------------Final Score-------------------------------")
|
||||
print(f"Final Score(Total 1000): {Final_score}")
|
||||
else:
|
||||
for i in range(len(data)):
|
||||
num_all[data[i]['dataset_name']] += 1
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
AllDataset_score[data[i]['dataset_name']] += data[i]['result']
|
||||
for key in AllDataset_score.keys():
|
||||
print(f"{key}: {AllDataset_score[key]/float(num_all[key])}")
|
||||
191
scripts/mPLUG-owl2.py
Normal file
191
scripts/mPLUG-owl2.py
Normal file
@@ -0,0 +1,191 @@
|
||||
import json
|
||||
from argparse import ArgumentParser
|
||||
import torch
|
||||
import os
|
||||
import json
|
||||
from tqdm import tqdm
|
||||
from PIL import Image
|
||||
import math
|
||||
import multiprocessing
|
||||
from multiprocessing import Pool, Queue, Manager
|
||||
|
||||
from transformers import TextStreamer
|
||||
from mplug_owl2.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN
|
||||
from mplug_owl2.conversation import conv_templates, SeparatorStyle
|
||||
from mplug_owl2.model.builder import load_pretrained_model
|
||||
from mplug_owl2.mm_utils import process_images, tokenizer_image_token, get_model_name_from_path, KeywordsStoppingCriteria
|
||||
|
||||
# https://github.com/X-PLUG/mPLUG-Owl/tree/main/mPLUG-Owl2
|
||||
def split_list(lst, n):
|
||||
length = len(lst)
|
||||
avg = length // n # 每份的大小
|
||||
result = [] # 存储分割后的子列表
|
||||
for i in range(n - 1):
|
||||
result.append(lst[i*avg:(i+1)*avg])
|
||||
result.append(lst[(n-1)*avg:])
|
||||
return result
|
||||
|
||||
def save_json(json_list,save_path):
|
||||
with open(save_path, 'w') as file:
|
||||
json.dump(json_list, file,indent=4)
|
||||
|
||||
def _get_args():
|
||||
parser = ArgumentParser()
|
||||
parser.add_argument("--image_folder", type=str, default="./data")
|
||||
parser.add_argument("--output_folder", type=str, default="./results")
|
||||
parser.add_argument("--OCRBench_file", type=str, default="./OCRBench/OCRBench.json")
|
||||
parser.add_argument("--model_path", type=str, default="./model_weights/mplug-owl2")
|
||||
parser.add_argument("--save_name", type=str, default="mplug-owl2")
|
||||
parser.add_argument("--conv_mode", type=str, default="mplug_owl2")
|
||||
parser.add_argument("--num_workers", type=int, default=8)
|
||||
parser.add_argument("--temperature", type=float, default=0.0)
|
||||
args = parser.parse_args()
|
||||
return args
|
||||
|
||||
OCRBench_score = {"Regular Text Recognition":0,"Irregular Text Recognition":0,"Artistic Text Recognition":0,"Handwriting Recognition":0,
|
||||
"Digit String Recognition":0,"Non-Semantic Text Recognition":0,"Scene Text-centric VQA":0,"Doc-oriented VQA":0,"Doc-oriented VQA":0,
|
||||
"Key Information Extraction":0,"Handwritten Mathematical Expression Recognition":0}
|
||||
AllDataset_score = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
num_all = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
|
||||
def eval_worker(args, data, eval_id, output_queue):
|
||||
print(f"Process {eval_id} start.")
|
||||
model_name = get_model_name_from_path(args.model_path)
|
||||
tokenizer, model, image_processor, context_len = load_pretrained_model(args.model_path, None, model_name, load_8bit=False, load_4bit=False, device=f"cuda:{eval_id}")
|
||||
for i in tqdm(range(len(data))):
|
||||
img_path = os.path.join(args.image_folder, data[i]['image_path'])
|
||||
qs = data[i]['question']
|
||||
if data[i].get("predict", 0)!=0:
|
||||
print(f"{img_path} predict exist, continue.")
|
||||
continue
|
||||
conv = conv_templates[args.conv_mode].copy()
|
||||
roles = conv.roles
|
||||
image = Image.open(img_path).convert('RGB')
|
||||
max_edge = max(image.size) # We recommand you to resize to squared image for BEST performance.
|
||||
image = image.resize((max_edge, max_edge))
|
||||
image_tensor = process_images([image], image_processor)
|
||||
image_tensor = image_tensor.to(model.device, dtype=torch.float16)
|
||||
|
||||
inp = DEFAULT_IMAGE_TOKEN + qs
|
||||
conv.append_message(conv.roles[0], inp)
|
||||
conv.append_message(conv.roles[1], None)
|
||||
prompt = conv.get_prompt()
|
||||
input_ids = tokenizer_image_token(prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors='pt').unsqueeze(0).to(model.device)
|
||||
stop_str = conv.sep2
|
||||
keywords = [stop_str]
|
||||
stopping_criteria = KeywordsStoppingCriteria(keywords, tokenizer, input_ids)
|
||||
streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
|
||||
with torch.inference_mode():
|
||||
output_ids = model.generate(
|
||||
input_ids,
|
||||
images=image_tensor,
|
||||
do_sample=False,
|
||||
temperature=args.temperature,
|
||||
max_new_tokens=100,
|
||||
streamer=streamer,
|
||||
use_cache=True,
|
||||
stopping_criteria=[stopping_criteria])
|
||||
outputs = tokenizer.decode(output_ids[0, input_ids.shape[1]:]).strip()
|
||||
data[i]['predict'] = outputs
|
||||
output_queue.put({eval_id: data})
|
||||
print(f"Process {eval_id} has completed.")
|
||||
|
||||
if __name__=="__main__":
|
||||
multiprocessing.set_start_method('spawn')
|
||||
args = _get_args()
|
||||
|
||||
if os.path.exists(os.path.join(args.output_folder,f"{args.save_name}.json")):
|
||||
data_path = os.path.join(args.output_folder,f"{args.save_name}.json")
|
||||
print(f"output_path:{data_path} exist! Only generate the results that were not generated in {data_path}.")
|
||||
else:
|
||||
data_path = args.OCRBench_file
|
||||
|
||||
with open(data_path, "r") as f:
|
||||
data = json.load(f)
|
||||
|
||||
data_list = split_list(data, args.num_workers)
|
||||
output_queue = Manager().Queue()
|
||||
|
||||
pool = Pool(processes=args.num_workers)
|
||||
for i in range(len(data_list)):
|
||||
pool.apply_async(eval_worker, args=(args, data_list[i], i, output_queue))
|
||||
pool.close()
|
||||
pool.join()
|
||||
|
||||
results = {}
|
||||
while not output_queue.empty():
|
||||
result = output_queue.get()
|
||||
results.update(result)
|
||||
data = []
|
||||
for i in range(len(data_list)):
|
||||
data.extend(results[i])
|
||||
|
||||
for i in range(len(data)):
|
||||
data_type = data[i]["type"]
|
||||
dataset_name = data[i]["dataset_name"]
|
||||
answers = data[i]["answers"]
|
||||
if data[i].get('predict',0)==0:
|
||||
continue
|
||||
predict = data[i]['predict']
|
||||
data[i]['result'] = 0
|
||||
if dataset_name == "HME100k":
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
save_json(data, os.path.join(args.output_folder,f"{args.save_name}.json"))
|
||||
if len(data)==1000:
|
||||
for i in range(len(data)):
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
OCRBench_score[data[i]['type']] += data[i]['result']
|
||||
recognition_score = OCRBench_score['Regular Text Recognition']+OCRBench_score['Irregular Text Recognition']+OCRBench_score['Artistic Text Recognition']+OCRBench_score['Handwriting Recognition']+OCRBench_score['Digit String Recognition']+OCRBench_score['Non-Semantic Text Recognition']
|
||||
Final_score = recognition_score+OCRBench_score['Scene Text-centric VQA']+OCRBench_score['Doc-oriented VQA']+OCRBench_score['Key Information Extraction']+OCRBench_score['Handwritten Mathematical Expression Recognition']
|
||||
print("###########################OCRBench##############################")
|
||||
print(f"Text Recognition(Total 300):{recognition_score}")
|
||||
print("------------------Details of Recognition Score-------------------")
|
||||
print(f"Regular Text Recognition(Total 50): {OCRBench_score['Regular Text Recognition']}")
|
||||
print(f"Irregular Text Recognition(Total 50): {OCRBench_score['Irregular Text Recognition']}")
|
||||
print(f"Artistic Text Recognition(Total 50): {OCRBench_score['Artistic Text Recognition']}")
|
||||
print(f"Handwriting Recognition(Total 50): {OCRBench_score['Handwriting Recognition']}")
|
||||
print(f"Digit String Recognition(Total 50): {OCRBench_score['Digit String Recognition']}")
|
||||
print(f"Non-Semantic Text Recognition(Total 50): {OCRBench_score['Non-Semantic Text Recognition']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Scene Text-centric VQA(Total 200): {OCRBench_score['Scene Text-centric VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Doc-oriented VQA(Total 200): {OCRBench_score['Doc-oriented VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Key Information Extraction(Total 200): {OCRBench_score['Key Information Extraction']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Handwritten Mathematical Expression Recognition(Total 100): {OCRBench_score['Handwritten Mathematical Expression Recognition']}")
|
||||
print("----------------------Final Score-------------------------------")
|
||||
print(f"Final Score(Total 1000): {Final_score}")
|
||||
else:
|
||||
for i in range(len(data)):
|
||||
num_all[data[i]['dataset_name']] += 1
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
AllDataset_score[data[i]['dataset_name']] += data[i]['result']
|
||||
for key in AllDataset_score.keys():
|
||||
print(f"{key}: {AllDataset_score[key]/float(num_all[key])}")
|
||||
187
scripts/minigpt4v2.py
Normal file
187
scripts/minigpt4v2.py
Normal file
@@ -0,0 +1,187 @@
|
||||
import json
|
||||
from argparse import ArgumentParser
|
||||
import torch
|
||||
import os
|
||||
import json
|
||||
from tqdm import tqdm
|
||||
from PIL import Image
|
||||
import math
|
||||
import multiprocessing
|
||||
from multiprocessing import Pool, Queue, Manager
|
||||
|
||||
import sys
|
||||
sys.path.append("./scripts/MiniGPT-4/")
|
||||
from minigpt4.common.eval_utils import prepare_texts, init_model, eval_parser
|
||||
from minigpt4.conversation.conversation import CONV_VISION_minigptv2
|
||||
from minigpt4.common.config import Config
|
||||
import random
|
||||
# https://github.com/Vision-CAIR/MiniGPT-4/blob/main/eval_scripts/eval_vqa.py
|
||||
|
||||
|
||||
def split_list(lst, n):
|
||||
length = len(lst)
|
||||
avg = length // n # 每份的大小
|
||||
result = [] # 存储分割后的子列表
|
||||
for i in range(n - 1):
|
||||
result.append(lst[i*avg:(i+1)*avg])
|
||||
result.append(lst[(n-1)*avg:])
|
||||
return result
|
||||
|
||||
def save_json(json_list,save_path):
|
||||
with open(save_path, 'w') as file:
|
||||
json.dump(json_list, file,indent=4)
|
||||
|
||||
def _get_args():
|
||||
parser = ArgumentParser()
|
||||
parser.add_argument("--image_folder", type=str, default="./data")
|
||||
parser.add_argument("--output_folder", type=str, default="./results")
|
||||
parser.add_argument("--OCRBench_file", type=str, default="./OCRBench/OCRBench.json")
|
||||
parser.add_argument("--cfg-path", default='./scripts/MiniGPT-4/eval_configs/minigptv2_eval.yaml')
|
||||
parser.add_argument("--save_name", type=str, default="minigptv2")
|
||||
parser.add_argument("--num_workers", type=int, default=1)
|
||||
parser.add_argument("--temperature", type=float, default=0.0)
|
||||
parser.add_argument(
|
||||
"--options",
|
||||
nargs="+",
|
||||
help="override some settings in the used config, the key-value pair "
|
||||
"in xxx=yyy format will be merged into config file (deprecate), "
|
||||
"change to --cfg-options instead.",
|
||||
)
|
||||
args = parser.parse_args()
|
||||
return args
|
||||
|
||||
OCRBench_score = {"Regular Text Recognition":0,"Irregular Text Recognition":0,"Artistic Text Recognition":0,"Handwriting Recognition":0,
|
||||
"Digit String Recognition":0,"Non-Semantic Text Recognition":0,"Scene Text-centric VQA":0,"Doc-oriented VQA":0,"Doc-oriented VQA":0,
|
||||
"Key Information Extraction":0,"Handwritten Mathematical Expression Recognition":0}
|
||||
AllDataset_score = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
num_all = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
|
||||
def eval_worker(args, data, eval_id, output_queue):
|
||||
|
||||
print(f"Process {eval_id} start.")
|
||||
device = f'cuda:{eval_id}'
|
||||
cfg = Config(args)
|
||||
model, vis_processor = init_model(args, device)
|
||||
conv_temp = CONV_VISION_minigptv2.copy()
|
||||
conv_temp.system = ""
|
||||
model.eval()
|
||||
instruction_pool = [
|
||||
"[vqa] {}"
|
||||
]
|
||||
for i in tqdm(range(len(data))):
|
||||
img_path = os.path.join(args.image_folder, data[i]['image_path'])
|
||||
qs = data[i]['question']
|
||||
if data[i].get("predict", 0)!=0:
|
||||
print(f"{img_path} predict exist, continue.")
|
||||
continue
|
||||
image = Image.open(img_path).convert("RGB")
|
||||
image = vis_processor(image)
|
||||
image = image.unsqueeze(0).to(device)
|
||||
# question = self.text_processor(qs)
|
||||
instruction = random.choice(instruction_pool).format(qs)
|
||||
instruction = "<Img><ImageHere></Img> {} ".format(instruction)
|
||||
texts = prepare_texts(instruction, conv_temp) # warp the texts with conversation template
|
||||
answers = model.generate(image, texts, max_new_tokens=100, do_sample=False)
|
||||
data[i]['predict'] = answers[0]
|
||||
output_queue.put({eval_id: data})
|
||||
print(f"Process {eval_id} has completed.")
|
||||
|
||||
if __name__=="__main__":
|
||||
multiprocessing.set_start_method('spawn')
|
||||
args = _get_args()
|
||||
|
||||
if os.path.exists(os.path.join(args.output_folder,f"{args.save_name}.json")):
|
||||
data_path = os.path.join(args.output_folder,f"{args.save_name}.json")
|
||||
print(f"output_path:{data_path} exist! Only generate the results that were not generated in {data_path}.")
|
||||
else:
|
||||
data_path = args.OCRBench_file
|
||||
|
||||
with open(data_path, "r") as f:
|
||||
data = json.load(f)
|
||||
|
||||
data_list = split_list(data, args.num_workers)
|
||||
output_queue = Manager().Queue()
|
||||
|
||||
pool = Pool(processes=args.num_workers)
|
||||
for i in range(len(data_list)):
|
||||
pool.apply_async(eval_worker, args=(args, data_list[i], i, output_queue))
|
||||
pool.close()
|
||||
pool.join()
|
||||
|
||||
results = {}
|
||||
while not output_queue.empty():
|
||||
result = output_queue.get()
|
||||
results.update(result)
|
||||
data = []
|
||||
for i in range(len(data_list)):
|
||||
data.extend(results[i])
|
||||
|
||||
for i in range(len(data)):
|
||||
data_type = data[i]["type"]
|
||||
dataset_name = data[i]["dataset_name"]
|
||||
answers = data[i]["answers"]
|
||||
if data[i].get('predict',0)==0:
|
||||
continue
|
||||
predict = data[i]['predict']
|
||||
data[i]['result'] = 0
|
||||
if dataset_name == "HME100k":
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
save_json(data, os.path.join(args.output_folder,f"{args.save_name}.json"))
|
||||
if len(data)==1000:
|
||||
for i in range(len(data)):
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
OCRBench_score[data[i]['type']] += data[i]['result']
|
||||
recognition_score = OCRBench_score['Regular Text Recognition']+OCRBench_score['Irregular Text Recognition']+OCRBench_score['Artistic Text Recognition']+OCRBench_score['Handwriting Recognition']+OCRBench_score['Digit String Recognition']+OCRBench_score['Non-Semantic Text Recognition']
|
||||
Final_score = recognition_score+OCRBench_score['Scene Text-centric VQA']+OCRBench_score['Doc-oriented VQA']+OCRBench_score['Key Information Extraction']+OCRBench_score['Handwritten Mathematical Expression Recognition']
|
||||
print("###########################OCRBench##############################")
|
||||
print(f"Text Recognition(Total 300):{recognition_score}")
|
||||
print("------------------Details of Recognition Score-------------------")
|
||||
print(f"Regular Text Recognition(Total 50): {OCRBench_score['Regular Text Recognition']}")
|
||||
print(f"Irregular Text Recognition(Total 50): {OCRBench_score['Irregular Text Recognition']}")
|
||||
print(f"Artistic Text Recognition(Total 50): {OCRBench_score['Artistic Text Recognition']}")
|
||||
print(f"Handwriting Recognition(Total 50): {OCRBench_score['Handwriting Recognition']}")
|
||||
print(f"Digit String Recognition(Total 50): {OCRBench_score['Digit String Recognition']}")
|
||||
print(f"Non-Semantic Text Recognition(Total 50): {OCRBench_score['Non-Semantic Text Recognition']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Scene Text-centric VQA(Total 200): {OCRBench_score['Scene Text-centric VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Doc-oriented VQA(Total 200): {OCRBench_score['Doc-oriented VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Key Information Extraction(Total 200): {OCRBench_score['Key Information Extraction']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Handwritten Mathematical Expression Recognition(Total 100): {OCRBench_score['Handwritten Mathematical Expression Recognition']}")
|
||||
print("----------------------Final Score-------------------------------")
|
||||
print(f"Final Score(Total 1000): {Final_score}")
|
||||
else:
|
||||
for i in range(len(data)):
|
||||
num_all[data[i]['dataset_name']] += 1
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
AllDataset_score[data[i]['dataset_name']] += data[i]['result']
|
||||
for key in AllDataset_score.keys():
|
||||
print(f"{key}: {AllDataset_score[key]/float(num_all[key])}")
|
||||
182
scripts/monkey.py
Normal file
182
scripts/monkey.py
Normal file
@@ -0,0 +1,182 @@
|
||||
import json
|
||||
from argparse import ArgumentParser
|
||||
import torch
|
||||
import os
|
||||
import json
|
||||
from tqdm import tqdm
|
||||
from PIL import Image
|
||||
import math
|
||||
import multiprocessing
|
||||
from multiprocessing import Pool, Queue, Manager
|
||||
|
||||
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||||
|
||||
# https://github.com/Yuliang-Liu/Monkey
|
||||
|
||||
def split_list(lst, n):
|
||||
length = len(lst)
|
||||
avg = length // n # 每份的大小
|
||||
result = [] # 存储分割后的子列表
|
||||
for i in range(n - 1):
|
||||
result.append(lst[i*avg:(i+1)*avg])
|
||||
result.append(lst[(n-1)*avg:])
|
||||
return result
|
||||
|
||||
def save_json(json_list,save_path):
|
||||
with open(save_path, 'w') as file:
|
||||
json.dump(json_list, file,indent=4)
|
||||
|
||||
def _get_args():
|
||||
parser = ArgumentParser()
|
||||
parser.add_argument("--image_folder", type=str, default="./data")
|
||||
parser.add_argument("--output_folder", type=str, default="./results")
|
||||
parser.add_argument("--OCRBench_file", type=str, default="./OCRBench/OCRBench.json")
|
||||
parser.add_argument("--model_path", type=str, default="echo840/Monkey")
|
||||
parser.add_argument("--save_name", type=str, default="monkey")
|
||||
parser.add_argument("--num_workers", type=int, default=8)
|
||||
args = parser.parse_args()
|
||||
return args
|
||||
|
||||
OCRBench_score = {"Regular Text Recognition":0,"Irregular Text Recognition":0,"Artistic Text Recognition":0,"Handwriting Recognition":0,
|
||||
"Digit String Recognition":0,"Non-Semantic Text Recognition":0,"Scene Text-centric VQA":0,"Doc-oriented VQA":0,"Doc-oriented VQA":0,
|
||||
"Key Information Extraction":0,"Handwritten Mathematical Expression Recognition":0}
|
||||
AllDataset_score = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
num_all = {"IIIT5K":0,"svt":0,"IC13_857":0,"IC15_1811":0,"svtp":0,"ct80":0,"cocotext":0,"ctw":0,"totaltext":0,"HOST":0,"WOST":0,"WordArt":0,"IAM":0,"ReCTS":0,"ORAND":0,"NonSemanticText":0,"SemanticText":0,
|
||||
"STVQA":0,"textVQA":0,"ocrVQA":0,"ESTVQA":0,"ESTVQA_cn":0,"docVQA":0,"infographicVQA":0,"ChartQA":0,"ChartQA_Human":0,"FUNSD":0,"SROIE":0,"POIE":0,"HME100k":0}
|
||||
|
||||
def eval_worker(args, data, eval_id, output_queue):
|
||||
print(f"Process {eval_id} start.")
|
||||
checkpoint = args.model_path
|
||||
model = AutoModelForCausalLM.from_pretrained(checkpoint, device_map='cuda', trust_remote_code=True).eval()
|
||||
tokenizer = AutoTokenizer.from_pretrained(checkpoint, trust_remote_code=True)
|
||||
tokenizer.padding_side = 'left'
|
||||
tokenizer.pad_token_id = tokenizer.eod_id
|
||||
|
||||
for i in tqdm(range(len(data))):
|
||||
img_path = os.path.join(args.image_folder, data[i]['image_path'])
|
||||
qs = data[i]['question']
|
||||
query = f'<img>{img_path}</img> {qs} Answer: '
|
||||
|
||||
input_ids = tokenizer(query, return_tensors='pt', padding='longest')
|
||||
attention_mask = input_ids.attention_mask
|
||||
input_ids = input_ids.input_ids
|
||||
|
||||
pred = model.generate(
|
||||
input_ids=input_ids.to(f'cuda:{eval_id}'),
|
||||
attention_mask=attention_mask.to(f'cuda:{eval_id}'),
|
||||
do_sample=False,
|
||||
num_beams=1,
|
||||
max_new_tokens=100,
|
||||
min_new_tokens=1,
|
||||
length_penalty=1,
|
||||
num_return_sequences=1,
|
||||
output_hidden_states=True,
|
||||
use_cache=True,
|
||||
pad_token_id=tokenizer.eod_id,
|
||||
eos_token_id=tokenizer.eod_id,
|
||||
)
|
||||
response = tokenizer.decode(pred[0][input_ids.size(1):].cpu(), skip_special_tokens=True).strip()
|
||||
data[i]['predict'] = response
|
||||
output_queue.put({eval_id: data})
|
||||
print(f"Process {eval_id} has completed.")
|
||||
|
||||
if __name__=="__main__":
|
||||
multiprocessing.set_start_method('spawn')
|
||||
args = _get_args()
|
||||
|
||||
if os.path.exists(os.path.join(args.output_folder,f"{args.save_name}.json")):
|
||||
data_path = os.path.join(args.output_folder,f"{args.save_name}.json")
|
||||
print(f"output_path:{data_path} exist! Only generate the results that were not generated in {data_path}.")
|
||||
else:
|
||||
data_path = args.OCRBench_file
|
||||
|
||||
with open(data_path, "r") as f:
|
||||
data = json.load(f)
|
||||
|
||||
data_list = split_list(data, args.num_workers)
|
||||
|
||||
output_queue = Manager().Queue()
|
||||
|
||||
pool = Pool(processes=args.num_workers)
|
||||
for i in range(len(data_list)):
|
||||
pool.apply_async(eval_worker, args=(args, data_list[i], i, output_queue))
|
||||
pool.close()
|
||||
pool.join()
|
||||
|
||||
|
||||
results = {}
|
||||
while not output_queue.empty():
|
||||
result = output_queue.get()
|
||||
results.update(result)
|
||||
data = []
|
||||
for i in range(len(data_list)):
|
||||
data.extend(results[i])
|
||||
|
||||
for i in range(len(data)):
|
||||
data_type = data[i]["type"]
|
||||
dataset_name = data[i]["dataset_name"]
|
||||
answers = data[i]["answers"]
|
||||
if data[i].get('predict',0)==0:
|
||||
continue
|
||||
predict = data[i]['predict']
|
||||
data[i]['result'] = 0
|
||||
if dataset_name == "HME100k":
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.strip().replace("\n"," ").replace(" ","")
|
||||
predict = predict.strip().replace("\n"," ").replace(" ","")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
if type(answers)==list:
|
||||
for j in range(len(answers)):
|
||||
answer = answers[j].lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answer in predict:
|
||||
data[i]['result'] = 1
|
||||
else:
|
||||
answers = answers.lower().strip().replace("\n"," ")
|
||||
predict = predict.lower().strip().replace("\n"," ")
|
||||
if answers in predict:
|
||||
data[i]['result'] = 1
|
||||
save_json(data, os.path.join(args.output_folder,f"{args.save_name}.json"))
|
||||
if len(data)==1000:
|
||||
for i in range(len(data)):
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
OCRBench_score[data[i]['type']] += data[i]['result']
|
||||
recognition_score = OCRBench_score['Regular Text Recognition']+OCRBench_score['Irregular Text Recognition']+OCRBench_score['Artistic Text Recognition']+OCRBench_score['Handwriting Recognition']+OCRBench_score['Digit String Recognition']+OCRBench_score['Non-Semantic Text Recognition']
|
||||
Final_score = recognition_score+OCRBench_score['Scene Text-centric VQA']+OCRBench_score['Doc-oriented VQA']+OCRBench_score['Key Information Extraction']+OCRBench_score['Handwritten Mathematical Expression Recognition']
|
||||
print("###########################OCRBench##############################")
|
||||
print(f"Text Recognition(Total 300):{recognition_score}")
|
||||
print("------------------Details of Recognition Score-------------------")
|
||||
print(f"Regular Text Recognition(Total 50): {OCRBench_score['Regular Text Recognition']}")
|
||||
print(f"Irregular Text Recognition(Total 50): {OCRBench_score['Irregular Text Recognition']}")
|
||||
print(f"Artistic Text Recognition(Total 50): {OCRBench_score['Artistic Text Recognition']}")
|
||||
print(f"Handwriting Recognition(Total 50): {OCRBench_score['Handwriting Recognition']}")
|
||||
print(f"Digit String Recognition(Total 50): {OCRBench_score['Digit String Recognition']}")
|
||||
print(f"Non-Semantic Text Recognition(Total 50): {OCRBench_score['Non-Semantic Text Recognition']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Scene Text-centric VQA(Total 200): {OCRBench_score['Scene Text-centric VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Doc-oriented VQA(Total 200): {OCRBench_score['Doc-oriented VQA']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Key Information Extraction(Total 200): {OCRBench_score['Key Information Extraction']}")
|
||||
print("----------------------------------------------------------------")
|
||||
print(f"Handwritten Mathematical Expression Recognition(Total 100): {OCRBench_score['Handwritten Mathematical Expression Recognition']}")
|
||||
print("----------------------Final Score-------------------------------")
|
||||
print(f"Final Score(Total 1000): {Final_score}")
|
||||
else:
|
||||
for i in range(len(data)):
|
||||
num_all[data[i]['dataset_name']] += 1
|
||||
if data[i].get("result",100)==100:
|
||||
continue
|
||||
AllDataset_score[data[i]['dataset_name']] += data[i]['result']
|
||||
for key in AllDataset_score.keys():
|
||||
print(f"{key}: {AllDataset_score[key]/float(num_all[key])}")
|
||||
Reference in New Issue
Block a user