exclude l1_loss in db model

odulcy-mindee · Feb 6, 2024 · c582655 · c582655
1 parent 114d7f0
commit c582655
Showing 1 changed file with 2 additions and 1 deletion.
diff --git a/doctr/models/detection/differentiable_binarization/pytorch.py b/doctr/models/detection/differentiable_binarization/pytorch.py
@@ -282,7 +282,8 @@ def compute_loss(
         if torch.any(thresh_mask):
             l1_loss = (torch.abs(thresh_map - thresh_target) * thresh_mask).sum() / (thresh_mask.sum() + eps)
 
-        return l1_loss + focal_scale * focal_loss + dice_loss
+        # return l1_loss + focal_scale * focal_loss + dice_loss
+        return focal_scale * focal_loss + dice_loss
 
 
 def _dbnet(