Add TV MaskRCNN Tile Recipe

src/otx/algo/instance_segmentation/maskrcnn_tv.py

@@ -56,6 +56,7 @@ def __init__(
             tile_config=tile_config,
         )
         self.image_size = (1, 3, 1024, 1024)
+        self.tile_image_size = (1, 3, 512, 512)
         self.mean = (123.675, 116.28, 103.53)
         self.std = (58.395, 57.12, 57.375)
 

src/otx/algo/instance_segmentation/torchvision/maskrcnn.py

@@ -36,11 +36,12 @@ def forward(
             entity.masks,
             entity.polygons,
         ):
+            # NOTE: shift labels by 1 as 0 is reserved for background
+            _labels = labels + 1 if len(labels) else labels
             targets.append(
                 {
                     "boxes": bboxes,
-                    # NOTE: shift labels by 1 as 0 is reserved for background
-                    "labels": labels + 1,
+                    "labels": _labels,
                     "masks": masks,
                     "polygons": polygons,
                 },

src/otx/recipe/instance_segmentation/maskrcnn_r50_tv_tile.yaml

@@ -0,0 +1,121 @@
+model:
+  class_path: otx.algo.instance_segmentation.maskrcnn_tv.TVMaskRCNNR50
+  init_args:
+    label_info: 80
+
+    optimizer:
+      class_path: torch.optim.SGD
+      init_args:
+        lr: 0.007
+        momentum: 0.9
+        weight_decay: 0.001
+
+    scheduler:
+      class_path: otx.core.schedulers.LinearWarmupSchedulerCallable
+      init_args:
+        num_warmup_steps: 100
+        main_scheduler_callable:
+          class_path: lightning.pytorch.cli.ReduceLROnPlateau
+          init_args:
+            mode: max
+            factor: 0.1
+            patience: 4
+            monitor: val/map_50
+
+engine:
+  task: INSTANCE_SEGMENTATION
+  device: auto
+
+callback_monitor: val/map_50
+
+data: ../_base_/data/torchvision_base.yaml
+overrides:
+  max_epochs: 100
+  gradient_clip_val: 35.0
+  data:
+    task: INSTANCE_SEGMENTATION
+    config:
+      stack_images: true
+      tile_config:
+        enable_tiler: true
+        enable_adaptive_tiling: true
+      data_format: coco_instances
+      include_polygons: true
+      train_subset:
+        batch_size: 4
+        num_workers: 8
+        transforms:
+          - class_path: otx.core.data.transform_libs.torchvision.Resize
+            init_args:
+              keep_ratio: false
+              transform_bbox: true
+              transform_mask: true
+              scale:
+                - 512
+                - 512
+          - class_path: otx.core.data.transform_libs.torchvision.Pad
+            init_args:
+              size_divisor: 32
+              transform_mask: true
+          - class_path: otx.core.data.transform_libs.torchvision.RandomFlip
+            init_args:
+              prob: 0.5
+              is_numpy_to_tvtensor: true
+          - class_path: torchvision.transforms.v2.ToDtype
+            init_args:
+              dtype: ${as_torch_dtype:torch.float32}
+              scale: False
+          - class_path: torchvision.transforms.v2.Normalize
+            init_args:
+              mean: [123.675, 116.28, 103.53]
+              std: [58.395, 57.12, 57.375]
+      val_subset:
+        batch_size: 1
+        num_workers: 4
+        transforms:
+          - class_path: otx.core.data.transform_libs.torchvision.Resize
+            init_args:
+              keep_ratio: false
+              transform_bbox: false
+              transform_mask: false
+              scale:
+                - 512
+                - 512
+          - class_path: otx.core.data.transform_libs.torchvision.Pad
+            init_args:
+              size_divisor: 32
+              transform_mask: false
+              is_numpy_to_tvtensor: true
+          - class_path: torchvision.transforms.v2.ToDtype
+            init_args:
+              dtype: ${as_torch_dtype:torch.float32}
+              scale: False
+          - class_path: torchvision.transforms.v2.Normalize
+            init_args:
+              mean: [123.675, 116.28, 103.53]
+              std: [58.395, 57.12, 57.375]
+      test_subset:
+        batch_size: 1
+        num_workers: 4
+        transforms:
+          - class_path: otx.core.data.transform_libs.torchvision.Resize
+            init_args:
+              keep_ratio: false
+              transform_bbox: false
+              transform_mask: false
+              scale:
+                - 512
+                - 512
+          - class_path: otx.core.data.transform_libs.torchvision.Pad
+            init_args:
+              size_divisor: 32
+              transform_mask: false
+              is_numpy_to_tvtensor: true
+          - class_path: torchvision.transforms.v2.ToDtype
+            init_args:
+              dtype: ${as_torch_dtype:torch.float32}
+              scale: False
+          - class_path: torchvision.transforms.v2.Normalize
+            init_args:
+              mean: [123.675, 116.28, 103.53]
+              std: [58.395, 57.12, 57.375]

tests/unit/algo/instance_segmentation/test_maskrcnn.py

@@ -5,6 +5,7 @@
 import pytest
 import torch
 from otx.algo.instance_segmentation.maskrcnn import MaskRCNNEfficientNet, MaskRCNNResNet50, MaskRCNNSwinT
+from otx.algo.instance_segmentation.maskrcnn_tv import TVMaskRCNNR50
 from otx.algo.utils.support_otx_v1 import OTXv1Helper
 from otx.core.data.entity.instance_segmentation import InstanceSegBatchPredEntity
 from otx.core.types.export import TaskLevelExportParameters
@@ -21,22 +22,29 @@ def test_load_weights(self, mocker) -> None:
 
     @pytest.mark.parametrize(
         "model",
-        [MaskRCNNResNet50(3), MaskRCNNEfficientNet(3), MaskRCNNSwinT(3)],
+        [MaskRCNNResNet50(3), MaskRCNNEfficientNet(3), MaskRCNNSwinT(3), TVMaskRCNNR50(3)],
     )
     def test_loss(self, model, fxt_data_module):
         data = next(iter(fxt_data_module.train_dataloader()))
         data.images = torch.randn([2, 3, 32, 32])
 
         output = model(data)
-        assert "loss_cls" in output
-        assert "loss_bbox" in output
-        assert "loss_mask" in output
-        assert "loss_rpn_cls" in output
-        assert "loss_rpn_bbox" in output
+        if isinstance(model, TVMaskRCNNR50):
+            assert "loss_classifier" in output
+            assert "loss_box_reg" in output
+            assert "loss_mask" in output
+            assert "loss_objectness" in output
+            assert "loss_rpn_box_reg" in output
+        else:
+            assert "loss_cls" in output
+            assert "loss_bbox" in output
+            assert "loss_mask" in output
+            assert "loss_rpn_cls" in output
+            assert "loss_rpn_bbox" in output
 
     @pytest.mark.parametrize(
         "model",
-        [MaskRCNNResNet50(3), MaskRCNNEfficientNet(3), MaskRCNNSwinT(3)],
+        [MaskRCNNResNet50(3), MaskRCNNEfficientNet(3), MaskRCNNSwinT(3), TVMaskRCNNR50(3)],
     )
     def test_predict(self, model, fxt_data_module):
         data = next(iter(fxt_data_module.train_dataloader()))
@@ -47,7 +55,7 @@ def test_predict(self, model, fxt_data_module):
 
     @pytest.mark.parametrize(
         "model",
-        [MaskRCNNResNet50(3), MaskRCNNEfficientNet(3), MaskRCNNSwinT(3)],
+        [MaskRCNNResNet50(3), MaskRCNNEfficientNet(3), MaskRCNNSwinT(3), TVMaskRCNNR50(3)],
     )
     def test_export(self, model):
         model.eval()