Spaces:
Running
on
A10G
Running
on
A10G
# Config for SmoothNet filter trained on Human3.6M data with a window size of | |
# 32. The model is trained using root-centered keypoint coordinates around the | |
# pelvis (index:0), thus we set root_index=0 for the filter | |
filter_cfg = dict( | |
type='SmoothNetFilter', | |
window_size=32, | |
output_size=32, | |
checkpoint='https://download.openmmlab.com/mmpose/plugin/smoothnet/' | |
'smoothnet_ws32_h36m.pth', | |
hidden_size=512, | |
res_hidden_size=256, | |
num_blocks=3, | |
root_index=0) | |