Add keypoint detection recipe for single object cases (#3903)

* add rtmpose_tiny for single obj

* add rtmpose_tiny for single obj

* modify test subset name

* fix unit test

* update recipe with reset

src/otx/algo/keypoint_detection/heads/rtmcc_head.py

@@ -192,7 +192,7 @@ def loss(self, x: tuple[Tensor], entity: KeypointDetBatchDataEntity) -> dict:
             mask=mask,
         )
 
-        loss_pose = torch.tensor(avg_acc, device=device)
+        loss_pose = -1 * torch.tensor(avg_acc, device=device)
         losses.update(loss_pose=loss_pose)
 
         return losses

src/otx/core/data/dataset/keypoint_detection.py

@@ -104,8 +104,8 @@ def _get_item_impl(self, index: int) -> KeypointDetDataEntity | None:
         ).reshape(-1, 2)
         keypoints_visible = np.minimum(1, keypoints)[..., 0]
 
-        bbox_center = (bboxes[0, 2:] + bboxes[0, :2]) * 0.5
-        bbox_scale = (bboxes[0, 2:] - bboxes[0, :2]) * 1.25
+        bbox_center = np.array(img_shape) / 2.0
+        bbox_scale = np.array(img_shape)
         bbox_rotation = 0.0
 
         entity = KeypointDetDataEntity(

src/otx/core/data/transform_libs/torchvision.py

@@ -3109,6 +3109,46 @@ def __call__(self, *_inputs: T_OTXDataEntity) -> T_OTXDataEntity | None:
         return inputs
 
 
+class GetBBoxCenterScale(tvt_v2.Transform):
+    """Convert bboxes from [x, y, w, h] to center and scale.
+
+    The center is the coordinates of the bbox center, and the scale is the
+    bbox width and height normalized by a scale factor.
+    Required Keys:
+        - bbox
+    Modified Keys:
+        - bbox_center
+        - bbox_scale
+    Args:
+        padding (float): The bbox padding scale that will be multilied to
+            `bbox_scale`. Defaults to 1.25
+    """
+
+    def __init__(self, padding: float = 1.25) -> None:
+        super().__init__()
+
+        self.padding = padding
+
+    def __call__(self, *_inputs: T_OTXDataEntity) -> T_OTXDataEntity | None:
+        """Transform function to add bbox_infos from bboxes for keypoint detection task."""
+        assert len(_inputs) == 1, "[tmp] Multiple entity is not supported yet."  # noqa: S101
+        inputs = _inputs[0]
+
+        bbox = inputs.bboxes[0].numpy()
+        inputs.bbox_info.center = (bbox[2:] + bbox[:2]) * 0.5
+        inputs.bbox_info.scale = (bbox[2:] - bbox[:2]) * self.padding
+
+        return inputs
+
+    def __repr__(self) -> str:
+        """Print the basic information of the transform.
+
+        Returns:
+            str: Formatted string.
+        """
+        return self.__class__.__name__ + f"(padding={self.padding})"
+
+
 class RandomBBoxTransform(tvt_v2.Transform):
     r"""Rnadomly shift, resize and rotate the bounding boxes.
 
@@ -3202,16 +3242,7 @@ def _get_transform_params(self) -> tuple:
         return offset, scale, rotate
 
     def __call__(self, *_inputs: T_OTXDataEntity) -> T_OTXDataEntity | None:
-        """The transform function of :class:`RandomBboxTransform`.
-
-        See ``transform()`` method of :class:`BaseTransform` for details.
-
-        Args:
-            results (dict): The result dict
-
-        Returns:
-            dict: The result dict.
-        """
+        """Transform function to adjust bbox_infos randomly."""
         assert len(_inputs) == 1, "[tmp] Multiple entity is not supported yet."  # noqa: S101
         inputs = _inputs[0]
 
@@ -3378,16 +3409,7 @@ def _get_warp_image(
         return torch.from_numpy(warped_image).permute(2, 0, 1)
 
     def __call__(self, *_inputs: T_OTXDataEntity) -> T_OTXDataEntity | None:
-        """The transform function of :class:`TopdownAffine`.
-
-        See ``transform()`` method of :class:`BaseTransform` for details.
-
-        Args:
-            results (dict): The result dict
-
-        Returns:
-            dict: The result dict.
-        """
+        """Transform function to affine image through warp matrix."""
         assert len(_inputs) == 1, "[tmp] Multiple entity is not supported yet."  # noqa: S101
         inputs = _inputs[0]
 

src/otx/recipe/_base_/data/keypoint_detection.yaml

@@ -12,6 +12,7 @@ train_subset:
   subset_name: train
   batch_size: 32
   transforms:
+    - class_path: otx.core.data.transform_libs.torchvision.GetBBoxCenterScale
     - class_path: otx.core.data.transform_libs.torchvision.RandomBBoxTransform
     - class_path: otx.core.data.transform_libs.torchvision.TopdownAffine
       init_args:
@@ -30,6 +31,7 @@ val_subset:
   subset_name: val
   batch_size: 32
   transforms:
+    - class_path: otx.core.data.transform_libs.torchvision.GetBBoxCenterScale
     - class_path: otx.core.data.transform_libs.torchvision.TopdownAffine
       init_args:
         input_size: $(input_size)
@@ -45,6 +47,7 @@ test_subset:
   subset_name: test
   batch_size: 32
   transforms:
+    - class_path: otx.core.data.transform_libs.torchvision.GetBBoxCenterScale
     - class_path: otx.core.data.transform_libs.torchvision.TopdownAffine
       init_args:
         input_size: $(input_size)

src/otx/recipe/keypoint_detection/rtmpose_tiny_single_obj.yaml

@@ -0,0 +1,81 @@
+model:
+  class_path: otx.algo.keypoint_detection.rtmpose.RTMPoseTiny
+  init_args:
+    label_info: 17
+
+    optimizer:
+      class_path: torch.optim.AdamW
+      init_args:
+        lr: 0.001
+        weight_decay: 0.0001
+
+    scheduler:
+      class_path: otx.core.schedulers.LinearWarmupSchedulerCallable
+      init_args:
+        num_warmup_steps: 3
+        main_scheduler_callable:
+          class_path: lightning.pytorch.cli.ReduceLROnPlateau
+          init_args:
+            mode: max
+            factor: 0.1
+            patience: 9
+            monitor: val/accuracy
+
+engine:
+  task: KEYPOINT_DETECTION
+  device: auto
+
+callback_monitor: val/accuracy
+
+data: ../_base_/data/keypoint_detection.yaml
+
+overrides:
+  gradient_clip_val: 35.0
+  reset:
+    - data.train_subset.transforms
+    - data.val_subset.transforms
+    - data.test_subset.transforms
+  input_size:
+    - 512
+    - 512
+  train_subset:
+    transforms:
+      - class_path: otx.core.data.transform_libs.torchvision.TopdownAffine
+        init_args:
+          input_size: $(input_size)
+      - class_path: otx.core.data.transform_libs.torchvision.YOLOXHSVRandomAug
+        init_args:
+          is_numpy_to_tvtensor: true
+      - class_path: torchvision.transforms.v2.ToDtype
+        init_args:
+          dtype: ${as_torch_dtype:torch.float32}
+      - class_path: torchvision.transforms.v2.Normalize
+        init_args:
+          mean: [123.675, 116.28, 103.53]
+          std: [58.395, 57.12, 57.375]
+  val_subset:
+    transforms:
+      - class_path: otx.core.data.transform_libs.torchvision.TopdownAffine
+        init_args:
+          input_size: $(input_size)
+          is_numpy_to_tvtensor: true
+      - class_path: torchvision.transforms.v2.ToDtype
+        init_args:
+          dtype: ${as_torch_dtype:torch.float32}
+      - class_path: torchvision.transforms.v2.Normalize
+        init_args:
+          mean: [123.675, 116.28, 103.53]
+          std: [58.395, 57.12, 57.375]
+  test_subset:
+    transforms:
+      - class_path: otx.core.data.transform_libs.torchvision.TopdownAffine
+        init_args:
+          input_size: $(input_size)
+          is_numpy_to_tvtensor: true
+      - class_path: torchvision.transforms.v2.ToDtype
+        init_args:
+          dtype: ${as_torch_dtype:torch.float32}
+      - class_path: torchvision.transforms.v2.Normalize
+        init_args:
+          mean: [123.675, 116.28, 103.53]
+          std: [58.395, 57.12, 57.375]

tests/unit/core/data/transform_libs/test_torchvision.py

@@ -20,8 +20,10 @@
 from otx.core.data.transform_libs.torchvision import (
     CachedMixUp,
     CachedMosaic,
+    Compose,
     DecodeVideo,
     FilterAnnotations,
+    GetBBoxCenterScale,
     MinIoURandomCrop,
     PackVideo,
     Pad,
@@ -901,12 +903,18 @@ def keypoint_det_entity(self) -> KeypointDetDataEntity:
             labels=torch.LongTensor([0]),
             keypoints=tv_tensors.TVTensor(np.array([[0, 4], [4, 2], [2, 6], [6, 0]])),
             keypoints_visible=tv_tensors.TVTensor(np.array([1, 1, 1, 0])),
-            bbox_info=BboxInfo(center=np.array([3.5, 3.5]), scale=np.array([8.75, 8.75]), rotation=0),
+            bbox_info=BboxInfo(center=np.array([5, 5]), scale=np.array([10, 10]), rotation=0),
         )
 
     def test_forward(self, keypoint_det_entity) -> None:
-        transform = TopdownAffine(input_size=(5, 5))
+        transform = Compose(
+            [
+                GetBBoxCenterScale(),
+                TopdownAffine(input_size=(5, 5)),
+            ],
+        )
         results = transform(deepcopy(keypoint_det_entity))
 
-        assert hasattr(results, "keypoints")
+        assert np.array_equal(results.bbox_info.center, np.array([3.5, 3.5]))
+        assert np.array_equal(results.bbox_info.scale, np.array([8.75, 8.75]))
         assert results.keypoints.shape == (4, 2)