Interhand 3d dataset (#468)

mmpose/core/evaluation/top_down_eval.py

@@ -12,14 +12,15 @@ def _calc_distances(preds, targets, mask, normalize):
     Note:
         batch_size: N
         num_keypoints: K
+        dimension of keypoints: D (normally, D=2 or D=3)
 
     Args:
-        preds (np.ndarray[N, K, 2]): Predicted keypoint location.
-        targets (np.ndarray[N, K, 2]): Groundtruth keypoint location.
+        preds (np.ndarray[N, K, D]): Predicted keypoint location.
+        targets (np.ndarray[N, K, D]): Groundtruth keypoint location.
         mask (np.ndarray[N, K]): Visibility of the target. False for invisible
             joints, and True for visible. Invisible joints will be ignored for
             accuracy calculation.
-        normalize (np.ndarray[N, 2]): Typical value is heatmap_size
+        normalize (np.ndarray[N, D]): Typical value is heatmap_size
 
     Returns:
         np.ndarray[K, N]: The normalized distances.
@@ -242,8 +243,10 @@ def keypoint_epe(pred, gt, mask):
     Returns:
         float: Average end-point error.
     """
+
     distances = _calc_distances(
-        pred, gt, mask, np.tile(np.array([[1, 1]]), (pred.shape[0], 1)))
+        pred, gt, mask,
+        np.ones((pred.shape[0], pred.shape[2]), dtype=np.float32))
     distance_valid = distances[distances != -1]
     return distance_valid.sum() / max(1, len(distance_valid))
 

mmpose/datasets/datasets/__init__.py

@@ -5,8 +5,8 @@
 from .face import (Face300WDataset, FaceAFLWDataset, FaceCOFWDataset,
                    FaceWFLWDataset)
 from .fashion import DeepFashionDataset
-from .hand import (FreiHandDataset, InterHand2DDataset, OneHand10KDataset,
-                   PanopticDataset)
+from .hand import (FreiHandDataset, InterHand2DDataset, InterHand3DDataset,
+                   OneHand10KDataset, PanopticDataset)
 from .mesh import (MeshAdversarialDataset, MeshH36MDataset, MeshMixDataset,
                    MoshDataset)
 from .top_down import (TopDownAicDataset, TopDownCocoDataset,
@@ -19,12 +19,12 @@
     'TopDownCocoDataset', 'BottomUpCocoDataset', 'BottomUpMhpDataset',
     'BottomUpAicDataset', 'TopDownMpiiDataset', 'TopDownMpiiTrbDataset',
     'OneHand10KDataset', 'PanopticDataset', 'FreiHandDataset',
-    'InterHand2DDataset', 'TopDownOCHumanDataset', 'TopDownAicDataset',
-    'TopDownCocoWholeBodyDataset', 'MeshH36MDataset', 'MeshMixDataset',
-    'MoshDataset', 'MeshAdversarialDataset', 'TopDownCrowdPoseDataset',
-    'BottomUpCrowdPoseDataset', 'TopDownFreiHandDataset',
-    'TopDownOneHand10KDataset', 'TopDownPanopticDataset',
-    'TopDownPoseTrack18Dataset', 'TopDownJhmdbDataset', 'TopDownMhpDataset',
-    'DeepFashionDataset', 'Face300WDataset', 'FaceAFLWDataset',
-    'FaceWFLWDataset', 'FaceCOFWDataset'
+    'InterHand2DDataset', 'InterHand3DDataset', 'TopDownOCHumanDataset',
+    'TopDownAicDataset', 'TopDownCocoWholeBodyDataset', 'MeshH36MDataset',
+    'MeshMixDataset', 'MoshDataset', 'MeshAdversarialDataset',
+    'TopDownCrowdPoseDataset', 'BottomUpCrowdPoseDataset',
+    'TopDownFreiHandDataset', 'TopDownOneHand10KDataset',
+    'TopDownPanopticDataset', 'TopDownPoseTrack18Dataset',
+    'TopDownJhmdbDataset', 'TopDownMhpDataset', 'DeepFashionDataset',
+    'Face300WDataset', 'FaceAFLWDataset', 'FaceWFLWDataset', 'FaceCOFWDataset'
 ]

mmpose/datasets/datasets/hand/__init__.py

@@ -1,9 +1,10 @@
 from .freihand_dataset import FreiHandDataset
 from .interhand2d_dataset import InterHand2DDataset
+from .interhand3d_dataset import InterHand3DDataset
 from .onehand10k_dataset import OneHand10KDataset
 from .panoptic_dataset import PanopticDataset
 
 __all__ = [
-    'FreiHandDataset', 'InterHand2DDataset', 'OneHand10KDataset',
-    'PanopticDataset'
+    'FreiHandDataset', 'InterHand2DDataset', 'InterHand3DDataset',
+    'OneHand10KDataset', 'PanopticDataset'
 ]