Update utils/model.py

utils/model.py

@@ -465,6 +465,9 @@ class OwlViTForClassification(nn.Module):
         print(f"im_features sum: {image_feats.sum().item()}, text_embeds sum: {text_embeds.sum().item()}")
         # Predict image-level classes (batch_size, num_patches, num_queries)
         image_text_logits, pred_logits, part_logits = self.cls_head(image_feats, text_embeds, topk_idxs)
+        print(f"topk_idxs: {topk_idxs}")
+        print(f"image_text_logits size: {image_text_logits.shape}")
+        print(f"image_text_logits sum: {image_text_logits.sum().item()}")
 
         if self.weight_dict["loss_xclip"] > 0:
             targets_cls = torch.tensor([target["targets_cls"] for target in targets]).unsqueeze(1).to(self.device)