what should i do if i want to evaluate ST-GCN on my dataset? about mmskeleton HOT 2 CLOSED

Hi, FesianXu,
You can directly employ the default feeder without any modification.
To generate the database (*data.npy and *label.pkl) for the feeder, please refer to
https://github.com/yysijie/st-gcn/blob/99163728404fcf536912d1e045cfb62eb544ab91/tools/ntu_gendata.py#L77

On your second question, we use the different method to process data in the different dataset.
In the NTU-RGB+D, we simply pad the sequences with zeros and ensure the final length of each sequence is equal.
In the Kinetics-skeleton, we random sample sub-sequence with the same window size. Please refer to
https://github.com/yysijie/st-gcn/blob/99163728404fcf536912d1e045cfb62eb544ab91/st_gcn/feeder/tools.py#L42

from mmskeleton.

Thanks for your response. It really helps.

from mmskeleton.