Open 112200tyh opened 4 months ago
Just modify the feeder_{dataset}.py in CTR-GCN to get frame-level sequence and add the corresponding field in config.yaml. For example, add these code in feeder_ntu.py and joint_frame.yaml
if self.frame:
if(data_numpy[:,:,:,1].sum()==0):
data_numpy[:,:,:,0] = data_numpy[:,:,:,0] - data_numpy[:, :, 1:2, 0]
else:
data_numpy = data_numpy - data_numpy[:, :,1:2,0:1]
if valid_frame_num != np.sum(data_numpy.sum(0).sum(-1).sum(-1) != 0):
print("valid_frame_num is different between s trans and frame trans---------")
2. Add frame file in joint_frame.yaml
```yaml
train_feeder_args:
data_path: data/ntu/NTU60_CS.npz
split: train
debug: False
random_choose: False
random_shift: False
random_move: False
window_size: 64
normalization: False
random_rot: True
p_interval: [0.5, 1]
vel: False
bone: False
frame: True
test_feeder_args:
data_path: data/ntu/NTU60_CS.npz
split: test
window_size: 64
p_interval: [0.95]
vel: False
bone: False
debug: False
frame: True
Thanks to the author for contributing. Is there any source code for the CTR-GCN model to use these two preprocessing methods?