Failed to obtain the results of the paper in H3.6m

MediaBrain-SJTU / SPGSN

The source codes of 'Skeleton-parted graph scattering networks for 3D human motion prediction'. ECCV 2022

MIT License

29 stars 2 forks source link

Failed to obtain the results of the paper in H3.6m #3

Open wangzhihao-coder opened 1 year ago

wangzhihao-coder commented 1 year ago

I have trained SPGSN on the prediction of both 400ms/1000ms in h3.6m. But the performance of the model is poor compared with the paper. Is it possible to release checkpoint or training configuration?