First draft

sam2/build_sam.py

@@ -11,6 +11,8 @@ from hydra import compose
 from hydra.utils import instantiate
 from omegaconf import OmegaConf
 
+from huggingface_hub import hf_hub_download
+
 
 def build_sam2(
     config_file,
@@ -76,6 +78,18 @@ def build_sam2_video_predictor(
     return model
 
 
+def build_sam2_hf(model_id, **kwargs):
+    config_file = hf_hub_download(repo_id=model_id, filename=f"{model_id}.yaml")
+    ckpt_path = hf_hub_download(repo_id=model_id, filename=f"{model_id}.pt")
+    return build_sam2_video_predictor(config_file=config_file, ckpt_path=ckpt_path, **kwargs)
+
+
+def build_sam2_video_predictor_hf(model_id, **kwargs):
+    config_file = hf_hub_download(repo_id=model_id, filename=f"{model_id}.yaml")
+    ckpt_path = hf_hub_download(repo_id=model_id, filename=f"{model_id}.pt")
+    return build_sam2_video_predictor(config_file=config_file, ckpt_path=ckpt_path, **kwargs)
+
+
 def _load_checkpoint(model, ckpt_path):
     if ckpt_path is not None:
         sd = torch.load(ckpt_path, map_location="cpu")["model"]

sam2/sam2_image_predictor.py

@@ -13,7 +13,7 @@ import torch
 from PIL.Image import Image
 
 from sam2.modeling.sam2_base import SAM2Base
-
+from sam2.build_sam import build_sam2_hf
 from sam2.utils.transforms import SAM2Transforms
 
 
@@ -62,6 +62,20 @@ class SAM2ImagePredictor:
             (64, 64),
         ]
 
+    def from_pretrained(model_id: str, **kwargs) -> "SAM2ImagePredictor":
+        """
+        Load a pretrained model from the Hugging Face model hub.
+
+        Arguments:
+          model_id (str): The Hugging Face repository ID.
+          **kwargs: Additional arguments to pass to the model constructor.
+
+        Returns:
+          (SAM2ImagePredictor): The loaded model.
+        """
+        sam_model = build_sam2_hf(model_id, **kwargs)
+        return SAM2ImagePredictor(sam_model)
+
     @torch.no_grad()
     def set_image(
         self,