1. fix warnings. \n 2. support CPU mode. \n 3. update README.

groundingdino/models/GroundingDINO/backbone/position_encoding.py

@@ -111,11 +111,11 @@ class PositionEmbeddingSineHW(nn.Module):
             x_embed = x_embed / (x_embed[:, :, -1:] + eps) * self.scale
 
         dim_tx = torch.arange(self.num_pos_feats, dtype=torch.float32, device=x.device)
-        dim_tx = self.temperatureW ** (2 * (dim_tx // 2) / self.num_pos_feats)
+        dim_tx = self.temperatureW ** (2 * (torch.div(dim_tx, 2, rounding_mode='floor')) / self.num_pos_feats)
         pos_x = x_embed[:, :, :, None] / dim_tx
 
         dim_ty = torch.arange(self.num_pos_feats, dtype=torch.float32, device=x.device)
-        dim_ty = self.temperatureH ** (2 * (dim_ty // 2) / self.num_pos_feats)
+        dim_ty = self.temperatureH ** (2 * (torch.div(dim_ty, 2, rounding_mode='floor')) / self.num_pos_feats)
         pos_y = y_embed[:, :, :, None] / dim_ty
 
         pos_x = torch.stack(

groundingdino/models/GroundingDINO/ms_deform_attn.py

@@ -25,7 +25,10 @@ from torch.autograd import Function
 from torch.autograd.function import once_differentiable
 from torch.nn.init import constant_, xavier_uniform_
 
-from groundingdino import _C
+try:
+    from groundingdino import _C
+except:
+    warnings.warn("Failed to load custom C++ ops. Running on CPU mode Only!")
 
 
 # helpers
@@ -323,6 +326,7 @@ class MultiScaleDeformableAttention(nn.Module):
                     reference_points.shape[-1]
                 )
             )
+    
         if torch.cuda.is_available() and value.is_cuda:
             halffloat = False
             if value.dtype == torch.float16:

groundingdino/models/GroundingDINO/utils.py

@@ -206,7 +206,7 @@ def gen_sineembed_for_position(pos_tensor):
     # sineembed_tensor = torch.zeros(n_query, bs, 256)
     scale = 2 * math.pi
     dim_t = torch.arange(128, dtype=torch.float32, device=pos_tensor.device)
-    dim_t = 10000 ** (2 * (dim_t // 2) / 128)
+    dim_t = 10000 ** (2 * (torch.div(dim_t, 2, rounding_mode='floor')) / 128)
     x_embed = pos_tensor[:, :, 0] * scale
     y_embed = pos_tensor[:, :, 1] * scale
     pos_x = x_embed[:, :, None] / dim_t