Initial snapshot: hand motion pipeline (Dyn-HaMR + dex-retargeting + SPIDER)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-09-15 14:13:49 +08:00
commit 7e4ef6f98b
7002 changed files with 1452249 additions and 0 deletions
@@ -0,0 +1,36 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import numpy as np
+
+from mmpose.core.post_processing.one_euro_filter import OneEuroFilter
+
+
+def test_one_euro_filter():
+    np.random.seed(1)
+
+    kpts = []
+    frames = 100
+    for i in range(frames):
+        kpts.append({
+            'keypoints': np.tile(np.array([10, 10, 0.9]), [17, 1]),
+            'area': 100,
+            'score': 0.9
+        })
+        kpts.append({
+            'keypoints': np.tile(np.array([11, 11, 0.9]), [17, 1]),
+            'area': 100,
+            'score': 0.8
+        })
+
+    one_euro_filter = OneEuroFilter(
+        kpts[0]['keypoints'][:, :2], min_cutoff=1.7, beta=0.3, fps=30)
+
+    for i in range(1, len(kpts)):
+        kpts[i]['keypoints'][:, :2] = one_euro_filter(
+            kpts[i]['keypoints'][:, :2])
+
+    one_euro_filter = OneEuroFilter(
+        kpts[0]['keypoints'][:, :2], min_cutoff=1.7, beta=0.3)
+
+    for i in range(1, len(kpts)):
+        kpts[i]['keypoints'][:, :2] = one_euro_filter(
+            kpts[i]['keypoints'][:, :2])
@@ -0,0 +1,72 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import torch
+
+from mmpose.core.post_processing.group import HeatmapParser
+
+
+def test_group():
+    cfg = {}
+    cfg['num_joints'] = 17
+    cfg['detection_threshold'] = 0.1
+    cfg['tag_threshold'] = 1
+    cfg['use_detection_val'] = True
+    cfg['ignore_too_much'] = False
+    cfg['nms_kernel'] = 5
+    cfg['nms_padding'] = 2
+    cfg['tag_per_joint'] = True
+    cfg['max_num_people'] = 1
+    parser = HeatmapParser(cfg)
+    fake_heatmap = torch.zeros(1, 1, 5, 5)
+    fake_heatmap[0, 0, 3, 3] = 1
+    fake_heatmap[0, 0, 3, 2] = 0.8
+    assert parser.nms(fake_heatmap)[0, 0, 3, 2] == 0
+    fake_heatmap = torch.zeros(1, 17, 32, 32)
+    fake_tag = torch.zeros(1, 17, 32, 32, 1)
+    fake_heatmap[0, 0, 10, 10] = 0.8
+    fake_heatmap[0, 1, 12, 12] = 0.9
+    fake_heatmap[0, 4, 8, 8] = 0.8
+    fake_heatmap[0, 8, 6, 6] = 0.9
+    fake_tag[0, 0, 10, 10] = 0.8
+    fake_tag[0, 1, 12, 12] = 0.9
+    fake_tag[0, 4, 8, 8] = 0.8
+    fake_tag[0, 8, 6, 6] = 0.9
+    grouped, scores = parser.parse(fake_heatmap, fake_tag, True, True)
+    assert grouped[0][0, 0, 0] == 10.25
+    assert abs(scores[0] - 0.2) < 0.001
+    cfg['tag_per_joint'] = False
+    parser = HeatmapParser(cfg)
+    grouped, scores = parser.parse(fake_heatmap, fake_tag, False, False)
+    assert grouped[0][0, 0, 0] == 10.
+    grouped, scores = parser.parse(fake_heatmap, fake_tag, False, True)
+    assert grouped[0][0, 0, 0] == 10.
+
+
+def test_group_score_per_joint():
+    cfg = {}
+    cfg['num_joints'] = 17
+    cfg['detection_threshold'] = 0.1
+    cfg['tag_threshold'] = 1
+    cfg['use_detection_val'] = True
+    cfg['ignore_too_much'] = False
+    cfg['nms_kernel'] = 5
+    cfg['nms_padding'] = 2
+    cfg['tag_per_joint'] = True
+    cfg['max_num_people'] = 1
+    cfg['score_per_joint'] = True
+    parser = HeatmapParser(cfg)
+    fake_heatmap = torch.zeros(1, 1, 5, 5)
+    fake_heatmap[0, 0, 3, 3] = 1
+    fake_heatmap[0, 0, 3, 2] = 0.8
+    assert parser.nms(fake_heatmap)[0, 0, 3, 2] == 0
+    fake_heatmap = torch.zeros(1, 17, 32, 32)
+    fake_tag = torch.zeros(1, 17, 32, 32, 1)
+    fake_heatmap[0, 0, 10, 10] = 0.8
+    fake_heatmap[0, 1, 12, 12] = 0.9
+    fake_heatmap[0, 4, 8, 8] = 0.8
+    fake_heatmap[0, 8, 6, 6] = 0.9
+    fake_tag[0, 0, 10, 10] = 0.8
+    fake_tag[0, 1, 12, 12] = 0.9
+    fake_tag[0, 4, 8, 8] = 0.8
+    fake_tag[0, 8, 6, 6] = 0.9
+    grouped, scores = parser.parse(fake_heatmap, fake_tag, True, True)
+    assert len(scores[0]) == 17
@@ -0,0 +1,81 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import numpy as np
+
+from mmpose.core.post_processing.nms import nms, oks_iou, oks_nms, soft_oks_nms
+
+
+def test_soft_oks_nms():
+    oks_thr = 0.9
+    kpts = []
+    kpts.append({
+        'keypoints': np.tile(np.array([10, 10, 0.9]), [17, 1]),
+        'area': 100,
+        'score': 0.9
+    })
+    kpts.append({
+        'keypoints': np.tile(np.array([10, 10, 0.9]), [17, 1]),
+        'area': 100,
+        'score': 0.4
+    })
+    kpts.append({
+        'keypoints': np.tile(np.array([100, 100, 0.9]), [17, 1]),
+        'area': 100,
+        'score': 0.7
+    })
+
+    keep = soft_oks_nms([kpts[i] for i in range(len(kpts))], oks_thr)
+    assert (keep == np.array([0, 2, 1])).all()
+
+    keep = oks_nms([kpts[i] for i in range(len(kpts))], oks_thr)
+    assert (keep == np.array([0, 2])).all()
+
+    kpts_with_score_joints = []
+    kpts_with_score_joints.append({
+        'keypoints':
+        np.tile(np.array([10, 10, 0.9]), [17, 1]),
+        'area':
+        100,
+        'score':
+        np.tile(np.array([0.9]), 17)
+    })
+    kpts_with_score_joints.append({
+        'keypoints':
+        np.tile(np.array([10, 10, 0.9]), [17, 1]),
+        'area':
+        100,
+        'score':
+        np.tile(np.array([0.4]), 17)
+    })
+    kpts_with_score_joints.append({
+        'keypoints':
+        np.tile(np.array([100, 100, 0.9]), [17, 1]),
+        'area':
+        100,
+        'score':
+        np.tile(np.array([0.7]), 17)
+    })
+    keep = soft_oks_nms([
+        kpts_with_score_joints[i] for i in range(len(kpts_with_score_joints))
+    ],
+                        oks_thr,
+                        score_per_joint=True)
+    assert (keep == np.array([0, 2, 1])).all()
+
+    keep = oks_nms([
+        kpts_with_score_joints[i] for i in range(len(kpts_with_score_joints))
+    ],
+                   oks_thr,
+                   score_per_joint=True)
+    assert (keep == np.array([0, 2])).all()
+
+
+def test_func_nms():
+    result = nms(np.array([[0, 0, 10, 10, 0.9], [0, 0, 10, 8, 0.8]]), 0.5)
+    assert result == [0]
+
+
+def test_oks_iou():
+    result = oks_iou(np.ones([17 * 3]), np.ones([1, 17 * 3]), 1, [1])
+    assert result[0] == 1.
+    result = oks_iou(np.zeros([17 * 3]), np.ones([1, 17 * 3]), 1, [1])
+    assert result[0] < 0.01