Fix empty bbox error for YOLOv9 (#4024)

* Fix empty bbox error

* Add unit test

* precommit

src/otx/algo/detection/losses/yolov9_loss.py

@@ -305,9 +305,20 @@ def __call__(self, target: Tensor, predict: tuple[Tensor, Tensor]) -> tuple[Tens
             predict (tuple[Tensor, Tensor]): The predicted class and bounding box.
 
         Returns:
-            tuple[Tensor, Tensor]: The aligned target tensor with (batch, targets, (class + 4)).
+            tuple[Tensor, Tensor]: The aligned target tensors with (batch, targets, (class + 4)) and (batch, targets).
         """
         predict_cls, predict_bbox = predict
+
+        # return if target has no gt information.
+        n_targets = target.shape[1]
+        if n_targets == 0:
+            device = predict_bbox.device
+            align_cls = torch.zeros_like(predict_cls, device=device)
+            align_bbox = torch.zeros_like(predict_bbox, device=device)
+            valid_mask = torch.zeros(predict_cls.shape[:2], dtype=bool, device=device)
+            anchor_matched_targets = torch.cat([align_cls, align_bbox], dim=-1)
+            return anchor_matched_targets, valid_mask
+
         target_cls, target_bbox = target.split([1, 4], dim=-1)  # B x N x (C B) -> B x N x C, B x N x B
         target_cls = target_cls.long().clamp(0)
 
@@ -341,8 +352,8 @@ def __call__(self, target: Tensor, predict: tuple[Tensor, Tensor]) -> tuple[Tens
         normalize_term = (target_matrix / (max_target + 1e-9)) * max_iou
         normalize_term = normalize_term.permute(0, 2, 1).gather(2, unique_indices)
         align_cls = align_cls * normalize_term * valid_mask[:, :, None]
-
-        return torch.cat([align_cls, align_bbox], dim=-1), valid_mask.bool()
+        anchor_matched_targets = torch.cat([align_cls, align_bbox], dim=-1)
+        return anchor_matched_targets, valid_mask
 
 
 class YOLOv9Criterion(nn.Module):

tests/unit/algo/detection/losses/test_yolov9_loss.py

@@ -163,6 +163,21 @@ def test_call(self, box_matcher: BoxMatcher) -> None:
         assert align_targets.shape == torch.Size([1, 3, 14])
         assert valid_masks.shape == torch.Size([1, 3])
 
+    def test_call_with_empty_bbox(self, box_matcher: BoxMatcher) -> None:
+        target = torch.zeros((1, 0, 5))
+
+        predict_cls = torch.rand((1, 8400, 10))
+        predict_bbox = torch.rand((1, 8400, 4))
+        predict = (predict_cls, predict_bbox)
+
+        align_targets, valid_masks = box_matcher(target, predict)
+
+        assert align_targets.shape == (1, 8400, 14)
+        assert torch.all(align_targets == 0)
+
+        assert valid_masks.shape == (1, 8400)
+        assert torch.all(~valid_masks)
+
 
 class TestYOLOv9Criterion:
     @pytest.fixture()