Upload 41 files

ntu120_xset/b_2/b_2.py

@@ -0,0 +1,98 @@
+modality = 'b'
+graph = 'nturgb+d'
+work_dir = './work_dirs/test_prototype/ntu120_xset/b_2'
+model = dict(
+    type='RecognizerGCN_7_1_1',
+    backbone=dict(
+        type='GCN_7_1_1',
+        tcn_ms_cfg=[(3, 1), (3, 2), (3, 3), (3, 4), ('max', 3), '1x1'],
+        graph_cfg=dict(
+            layout='nturgb+d',
+            mode='random',
+            num_filter=8,
+            init_off=0.04,
+            init_std=0.02)),
+    cls_head=dict(type='SimpleHead_7_4_3', num_classes=120, in_channels=384))
+dataset_type = 'PoseDataset'
+ann_file = '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl'
+train_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='RandomRot', theta=0.2),
+    dict(type='Spatial_Flip', dataset='nturgb+d', p=0.5),
+    dict(type='GenSkeFeat', feats=['b']),
+    dict(type='UniformSampleDecode', clip_len=100),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+val_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['b']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+test_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['b']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+data = dict(
+    videos_per_gpu=16,
+    workers_per_gpu=4,
+    test_dataloader=dict(videos_per_gpu=1),
+    train=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='RandomRot', theta=0.2),
+            dict(type='Spatial_Flip', dataset='nturgb+d', p=0.5),
+            dict(type='GenSkeFeat', feats=['b']),
+            dict(type='UniformSampleDecode', clip_len=100),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_train'),
+    val=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['b']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'),
+    test=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['b']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'))
+optimizer = dict(
+    type='SGD', lr=0.025, momentum=0.9, weight_decay=0.0005, nesterov=True)
+optimizer_config = dict(grad_clip=None)
+lr_config = dict(policy='CosineAnnealing', min_lr=0, by_epoch=False)
+total_epochs = 150
+checkpoint_config = dict(interval=1)
+evaluation = dict(interval=1, metrics=['top_k_accuracy'])
+log_config = dict(interval=100, hooks=[dict(type='TextLoggerHook')])
+dist_params = dict(backend='nccl')
+gpu_ids = range(0, 1)

ntu120_xset/b_2/best_pred.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dfaa02fa3fb1bc5134b5b0f6de9219975e409c9f041d4226192922507ace2990
+size 43573230

ntu120_xset/b_2/best_top1_acc_epoch_146.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e13da8ef8b5f82da4828d0d91a69e9e2ef2863c81164bea7eadb096576d36a7
+size 33657446

ntu120_xset/bm/best_pred.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84e82de920e8e276b637c9dd2d61271cfc4cbea0622a11a0e779802a9f89805b
+size 43572354

ntu120_xset/bm/best_top1_acc_epoch_143.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99b3cdcf1a5deb1f0181dcb2d4d6dc93535858667c993a4be2ffdaba3230fd44
+size 33657446

ntu120_xset/bm/bm.py

@@ -0,0 +1,98 @@
+modality = 'bm'
+graph = 'nturgb+d'
+work_dir = './work_dirs/test_prototype/ntu120_xset/bm'
+model = dict(
+    type='RecognizerGCN_7_1_1',
+    backbone=dict(
+        type='GCN_7_1_1',
+        tcn_ms_cfg=[(3, 1), (3, 2), (3, 3), (3, 4), ('max', 3), '1x1'],
+        graph_cfg=dict(
+            layout='nturgb+d',
+            mode='random',
+            num_filter=8,
+            init_off=0.04,
+            init_std=0.02)),
+    cls_head=dict(type='SimpleHead_7_4_3', num_classes=120, in_channels=384))
+dataset_type = 'PoseDataset'
+ann_file = '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl'
+train_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='RandomRot', theta=0.2),
+    dict(type='Spatial_Flip', dataset='nturgb+d', p=0.5),
+    dict(type='GenSkeFeat', feats=['bm']),
+    dict(type='UniformSampleDecode', clip_len=100),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+val_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['bm']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+test_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['bm']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+data = dict(
+    videos_per_gpu=16,
+    workers_per_gpu=4,
+    test_dataloader=dict(videos_per_gpu=1),
+    train=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='RandomRot', theta=0.2),
+            dict(type='Spatial_Flip', dataset='nturgb+d', p=0.5),
+            dict(type='GenSkeFeat', feats=['bm']),
+            dict(type='UniformSampleDecode', clip_len=100),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_train'),
+    val=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['bm']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'),
+    test=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['bm']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'))
+optimizer = dict(
+    type='SGD', lr=0.025, momentum=0.9, weight_decay=0.0005, nesterov=True)
+optimizer_config = dict(grad_clip=None)
+lr_config = dict(policy='CosineAnnealing', min_lr=0, by_epoch=False)
+total_epochs = 150
+checkpoint_config = dict(interval=1)
+evaluation = dict(interval=1, metrics=['top_k_accuracy'])
+log_config = dict(interval=100, hooks=[dict(type='TextLoggerHook')])
+dist_params = dict(backend='nccl')
+gpu_ids = range(0, 1)

ntu120_xset/j_1/best_pred.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:126365b090ae8a5eac0cf90455a29a069a0d6cba522394fefef5351aeffb7c27
+size 43571998

ntu120_xset/j_1/best_top1_acc_epoch_148.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4dab6f329ff8cb3a401402e51bd9c8be0f0ae0e652e0015ab100fa281205b50d
+size 35500646

ntu120_xset/j_1/j_1.py

@@ -0,0 +1,96 @@
+modality = 'j'
+graph = 'nturgb+d'
+work_dir = './work_dirs/test_prototype/ntu120_xset/j_1'
+model = dict(
+    type='RecognizerGCN_7_1_1',
+    backbone=dict(
+        type='GCN_7_1_2',
+        tcn_ms_cfg=[(3, 1), (3, 2), (3, 3), (3, 4), ('max', 3), '1x1'],
+        graph_cfg=dict(
+            layout='nturgb+d',
+            mode='random',
+            num_filter=8,
+            init_off=0.04,
+            init_std=0.02)),
+    cls_head=dict(type='SimpleHead_7_4_2', num_classes=120, in_channels=384))
+dataset_type = 'PoseDataset'
+ann_file = '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl'
+train_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='RandomRot', theta=0.2),
+    dict(type='GenSkeFeat', feats=['j']),
+    dict(type='UniformSampleDecode', clip_len=100),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+val_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['j']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+test_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['j']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+data = dict(
+    videos_per_gpu=16,
+    workers_per_gpu=4,
+    test_dataloader=dict(videos_per_gpu=1),
+    train=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='RandomRot', theta=0.2),
+            dict(type='GenSkeFeat', feats=['j']),
+            dict(type='UniformSampleDecode', clip_len=100),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_train'),
+    val=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['j']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'),
+    test=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['j']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'))
+optimizer = dict(
+    type='SGD', lr=0.025, momentum=0.9, weight_decay=0.0005, nesterov=True)
+optimizer_config = dict(grad_clip=None)
+lr_config = dict(policy='CosineAnnealing', min_lr=0, by_epoch=False)
+total_epochs = 150
+checkpoint_config = dict(interval=1)
+evaluation = dict(interval=1, metrics=['top_k_accuracy'])
+log_config = dict(interval=100, hooks=[dict(type='TextLoggerHook')])
+dist_params = dict(backend='nccl')
+gpu_ids = range(0, 1)

ntu120_xset/j_2/best_pred.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3377e356f45a027015f662cc8e603d0888c734c96606f538344510640ed7d175
+size 43568628

ntu120_xset/j_2/best_top1_acc_epoch_150.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:59021245430269bf2817118569fa1a227af07105c6bfccf0a17a7ebc3bb0e304
+size 35500646

ntu120_xset/j_2/j_2.py

@@ -0,0 +1,96 @@
+modality = 'j'
+graph = 'nturgb+d'
+work_dir = './work_dirs/test_prototype/ntu120_xset/j_2'
+model = dict(
+    type='RecognizerGCN_7_1_1',
+    backbone=dict(
+        type='GCN_7_1_2',
+        tcn_ms_cfg=[(3, 1), (3, 2), (3, 3), (3, 4), ('max', 3), '1x1'],
+        graph_cfg=dict(
+            layout='nturgb+d',
+            mode='random',
+            num_filter=8,
+            init_off=0.04,
+            init_std=0.02)),
+    cls_head=dict(type='SimpleHead_7_4_2', num_classes=120, in_channels=384))
+dataset_type = 'PoseDataset'
+ann_file = '/data1/hao.wang/reproducation/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl'
+train_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='RandomRot', theta=0.2),
+    dict(type='GenSkeFeat', feats=['j']),
+    dict(type='UniformSampleDecode', clip_len=100),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+val_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['j']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+test_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['j']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+data = dict(
+    videos_per_gpu=16,
+    workers_per_gpu=4,
+    test_dataloader=dict(videos_per_gpu=1),
+    train=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/hao.wang/reproducation/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='RandomRot', theta=0.2),
+            dict(type='GenSkeFeat', feats=['j']),
+            dict(type='UniformSampleDecode', clip_len=100),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_train'),
+    val=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/hao.wang/reproducation/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['j']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'),
+    test=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/hao.wang/reproducation/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['j']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'))
+optimizer = dict(
+    type='SGD', lr=0.025, momentum=0.9, weight_decay=0.0005, nesterov=True)
+optimizer_config = dict(grad_clip=None)
+lr_config = dict(policy='CosineAnnealing', min_lr=0, by_epoch=False)
+total_epochs = 150
+checkpoint_config = dict(interval=1)
+evaluation = dict(interval=1, metrics=['top_k_accuracy'])
+log_config = dict(interval=100, hooks=[dict(type='TextLoggerHook')])
+dist_params = dict(backend='nccl')
+gpu_ids = range(0, 1)

ntu120_xset/jm/best_pred.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef6c9719f76e4b2b071859717826ce53f3bb9683bf4ffca7b2183b6f373e7128
+size 43572115

ntu120_xset/jm/best_top1_acc_epoch_150.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:41c426eab0e26b5dfda66b4a16d6a1e1a771e3f13052f181f6ad33531ca1f802
+size 35500646

ntu120_xset/jm/jm.py

@@ -0,0 +1,96 @@
+modality = 'jm'
+graph = 'nturgb+d'
+work_dir = './work_dirs/test_prototype/ntu120_xset/jm'
+model = dict(
+    type='RecognizerGCN_7_1_1',
+    backbone=dict(
+        type='GCN_7_1_2',
+        tcn_ms_cfg=[(3, 1), (3, 2), (3, 3), (3, 4), ('max', 3), '1x1'],
+        graph_cfg=dict(
+            layout='nturgb+d',
+            mode='random',
+            num_filter=8,
+            init_off=0.04,
+            init_std=0.02)),
+    cls_head=dict(type='SimpleHead_7_4_2', num_classes=120, in_channels=384))
+dataset_type = 'PoseDataset'
+ann_file = '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl'
+train_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='RandomRot', theta=0.2),
+    dict(type='GenSkeFeat', feats=['jm']),
+    dict(type='UniformSampleDecode', clip_len=100),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+val_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['jm']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+test_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['jm']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+data = dict(
+    videos_per_gpu=16,
+    workers_per_gpu=4,
+    test_dataloader=dict(videos_per_gpu=1),
+    train=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='RandomRot', theta=0.2),
+            dict(type='GenSkeFeat', feats=['jm']),
+            dict(type='UniformSampleDecode', clip_len=100),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_train'),
+    val=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['jm']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'),
+    test=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['jm']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'))
+optimizer = dict(
+    type='SGD', lr=0.025, momentum=0.9, weight_decay=0.0005, nesterov=True)
+optimizer_config = dict(grad_clip=None)
+lr_config = dict(policy='CosineAnnealing', min_lr=0, by_epoch=False)
+total_epochs = 150
+checkpoint_config = dict(interval=1)
+evaluation = dict(interval=1, metrics=['top_k_accuracy'])
+log_config = dict(interval=100, hooks=[dict(type='TextLoggerHook')])
+dist_params = dict(backend='nccl')
+gpu_ids = range(0, 1)

ntu120_xset/k_1/best_pred.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f8239ec803a9d92f6a6f02e89e2990ed5d4dd271498adcc33e258151f221fcbe
+size 43579216

ntu120_xset/k_1/best_top1_acc_epoch_147.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:73b983ca1c1a18c12c55d3c0f087bdb94d18b0a2d09f72ca5484fec40530ffe6
+size 33657446

ntu120_xset/k_1/k_1.py

@@ -0,0 +1,98 @@
+modality = 'k'
+graph = 'nturgb+d'
+work_dir = './work_dirs/test_prototype/ntu120_xset/k_1'
+model = dict(
+    type='RecognizerGCN_7_1_1',
+    backbone=dict(
+        type='GCN_7_1_1',
+        tcn_ms_cfg=[(3, 1), (3, 2), (3, 3), (3, 4), ('max', 3), '1x1'],
+        graph_cfg=dict(
+            layout='nturgb+d',
+            mode='random',
+            num_filter=8,
+            init_off=0.04,
+            init_std=0.02)),
+    cls_head=dict(type='SimpleHead_7_4_1', num_classes=120, in_channels=384))
+dataset_type = 'PoseDataset'
+ann_file = '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl'
+train_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='RandomRot', theta=0.2),
+    dict(type='Spatial_Flip', dataset='nturgb+d', p=0.5),
+    dict(type='GenSkeFeat', feats=['k']),
+    dict(type='UniformSampleDecode', clip_len=100),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+val_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['k']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+test_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['k']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+data = dict(
+    videos_per_gpu=16,
+    workers_per_gpu=4,
+    test_dataloader=dict(videos_per_gpu=1),
+    train=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='RandomRot', theta=0.2),
+            dict(type='Spatial_Flip', dataset='nturgb+d', p=0.5),
+            dict(type='GenSkeFeat', feats=['k']),
+            dict(type='UniformSampleDecode', clip_len=100),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_train'),
+    val=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['k']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'),
+    test=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['k']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'))
+optimizer = dict(
+    type='SGD', lr=0.025, momentum=0.9, weight_decay=0.0005, nesterov=True)
+optimizer_config = dict(grad_clip=None)
+lr_config = dict(policy='CosineAnnealing', min_lr=0, by_epoch=False)
+total_epochs = 150
+checkpoint_config = dict(interval=1)
+evaluation = dict(interval=1, metrics=['top_k_accuracy'])
+log_config = dict(interval=100, hooks=[dict(type='TextLoggerHook')])
+dist_params = dict(backend='nccl')
+gpu_ids = range(0, 1)

ntu120_xset/k_2/best_pred.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b1d24c3519eada59fee1212973ae976667120a3cafae0c030cfb4b568f6e76c
+size 43571933

ntu120_xset/k_2/best_top1_acc_epoch_141.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:762a2f5c5e3da3575ee54d08625a4aa4e6bf9fb8ec2fb8920c73fe81be791962
+size 33657446

ntu120_xset/k_2/k_2.py

@@ -0,0 +1,98 @@
+modality = 'k'
+graph = 'nturgb+d'
+work_dir = './work_dirs/test_prototype/ntu120_xset/k_2'
+model = dict(
+    type='RecognizerGCN_7_1_1',
+    backbone=dict(
+        type='GCN_7_1_1',
+        tcn_ms_cfg=[(3, 1), (3, 2), (3, 3), (3, 4), ('max', 3), '1x1'],
+        graph_cfg=dict(
+            layout='nturgb+d',
+            mode='random',
+            num_filter=8,
+            init_off=0.04,
+            init_std=0.02)),
+    cls_head=dict(type='SimpleHead_7_4_1', num_classes=120, in_channels=384))
+dataset_type = 'PoseDataset'
+ann_file = '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl'
+train_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='RandomRot', theta=0.2),
+    dict(type='Spatial_Flip', dataset='nturgb+d', p=0.5),
+    dict(type='GenSkeFeat', feats=['k']),
+    dict(type='UniformSampleDecode', clip_len=100),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+val_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['k']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+test_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['k']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+data = dict(
+    videos_per_gpu=16,
+    workers_per_gpu=4,
+    test_dataloader=dict(videos_per_gpu=1),
+    train=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='RandomRot', theta=0.2),
+            dict(type='Spatial_Flip', dataset='nturgb+d', p=0.5),
+            dict(type='GenSkeFeat', feats=['k']),
+            dict(type='UniformSampleDecode', clip_len=100),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_train'),
+    val=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['k']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'),
+    test=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['k']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'))
+optimizer = dict(
+    type='SGD', lr=0.025, momentum=0.9, weight_decay=0.0005, nesterov=True)
+optimizer_config = dict(grad_clip=None)
+lr_config = dict(policy='CosineAnnealing', min_lr=0, by_epoch=False)
+total_epochs = 150
+checkpoint_config = dict(interval=1)
+evaluation = dict(interval=1, metrics=['top_k_accuracy'])
+log_config = dict(interval=100, hooks=[dict(type='TextLoggerHook')])
+dist_params = dict(backend='nccl')
+gpu_ids = range(0, 1)

ntu120_xset/km/best_pred.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f7636d736c05e624059b72783e8beca8ee8ba9c2539b40663e1e37babf8a783
+size 43570624

ntu120_xset/km/best_top1_acc_epoch_145.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2833e283aa25a0a74612a856744e8d9504c71bb93c0678c7f6cc92a7c1a6f0f
+size 33657446

ntu120_xset/km/km.py

@@ -0,0 +1,98 @@
+modality = 'km'
+graph = 'nturgb+d'
+work_dir = './work_dirs/test_prototype/ntu120_xset/km'
+model = dict(
+    type='RecognizerGCN_7_1_1',
+    backbone=dict(
+        type='GCN_7_1_1',
+        tcn_ms_cfg=[(3, 1), (3, 2), (3, 3), (3, 4), ('max', 3), '1x1'],
+        graph_cfg=dict(
+            layout='nturgb+d',
+            mode='random',
+            num_filter=8,
+            init_off=0.04,
+            init_std=0.02)),
+    cls_head=dict(type='SimpleHead_7_4_1', num_classes=120, in_channels=384))
+dataset_type = 'PoseDataset'
+ann_file = '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl'
+train_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='RandomRot', theta=0.2),
+    dict(type='Spatial_Flip', dataset='nturgb+d', p=0.5),
+    dict(type='GenSkeFeat', feats=['km']),
+    dict(type='UniformSampleDecode', clip_len=100),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+val_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['km']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+test_pipeline = [
+    dict(type='PreNormalize3D', align_spine=False),
+    dict(type='GenSkeFeat', feats=['km']),
+    dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+    dict(type='FormatGCNInput'),
+    dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['keypoint'])
+]
+data = dict(
+    videos_per_gpu=16,
+    workers_per_gpu=4,
+    test_dataloader=dict(videos_per_gpu=1),
+    train=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='RandomRot', theta=0.2),
+            dict(type='Spatial_Flip', dataset='nturgb+d', p=0.5),
+            dict(type='GenSkeFeat', feats=['km']),
+            dict(type='UniformSampleDecode', clip_len=100),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_train'),
+    val=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['km']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=1),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'),
+    test=dict(
+        type='PoseDataset',
+        ann_file=
+        '/data1/medical/hao.wang/reproduction/hongda.liu/pyskl_data/nturgbd/ntu120_3danno.pkl',
+        pipeline=[
+            dict(type='PreNormalize3D', align_spine=False),
+            dict(type='GenSkeFeat', feats=['km']),
+            dict(type='UniformSampleDecode', clip_len=100, num_clips=10),
+            dict(type='FormatGCNInput'),
+            dict(type='Collect', keys=['keypoint', 'label'], meta_keys=[]),
+            dict(type='ToTensor', keys=['keypoint'])
+        ],
+        split='xset_val'))
+optimizer = dict(
+    type='SGD', lr=0.025, momentum=0.9, weight_decay=0.0005, nesterov=True)
+optimizer_config = dict(grad_clip=None)
+lr_config = dict(policy='CosineAnnealing', min_lr=0, by_epoch=False)
+total_epochs = 150
+checkpoint_config = dict(interval=1)
+evaluation = dict(interval=1, metrics=['top_k_accuracy'])
+log_config = dict(interval=100, hooks=[dict(type='TextLoggerHook')])
+dist_params = dict(backend='nccl')
+gpu_ids = range(0, 1)

ntu120_xset/ntu120_xset_ensemble.py

@@ -0,0 +1,65 @@
+from mmcv import load
+import sys
+# Note: please adjust the relative path according to the actual situation.
+sys.path.append('../..')
+from protogcn.smp import *
+
+
+j_1 = load('j_1/best_pred.pkl')
+b_1 = load('b_1/best_pred.pkl')
+k_1 = load('k_1/best_pred.pkl')
+j_2 = load('j_2/best_pred.pkl')
+b_2 = load('b_2/best_pred.pkl')
+k_2 = load('k_2/best_pred.pkl')
+jm = load('jm/best_pred.pkl')
+bm = load('bm/best_pred.pkl')
+km = load('km/best_pred.pkl')
+label = load_label('/data/nturgbd/ntu120_3danno.pkl', 'xset_val')
+
+
+"""
+***************
+InfoGCN  v0:
+j jm b bm k km
+2S:     91.23
+4S:     91.51
+6S:     91.89
+***************
+"""
+print('InfoGCN  v0:')
+print('j jm b bm k km')
+print('2S')
+fused = comb([j_1, b_1], [1, 1])
+print('Top-1', top1(fused, label))
+
+print('4S')
+fused = comb([j_1, b_1, jm, bm], [3, 3, 1, 1])
+print('Top-1', top1(fused, label))
+
+print('6S')
+fused = comb([j_1, b_1, k_1, jm, bm, km], [3, 3, 3, 1, 1, 1])
+print('Top-1', top1(fused, label))
+
+
+"""
+***************
+InfoGCN  v1:
+j j b b k k
+2S:     91.23
+4S:     91.86
+6S:     92.16
+***************
+"""
+print('InfoGCN  v1:')
+print('j j b b k k')
+print('2S')
+fused = comb([j_1, b_1], [1, 1])
+print('Top-1', top1(fused, label))
+
+print('4S')
+fused = comb([j_1, b_1, j_2, b_2], [1, 1, 1, 1])
+print('Top-1', top1(fused, label))
+
+print('6S')
+fused = comb([j_1, j_2, b_1, b_2, k_1, k_2], [7, 7, 8, 8, 5, 5])
+print('Top-1', top1(fused, label))