I trained the ntu120 data joint mode on X-sub according to README.md, but the accuracy is only in the 44%, is this because there is something missing in the code?

damNull / LAGCN

41 stars 7 forks source link

I trained the ntu120 data joint mode on X-sub according to README.md, but the accuracy is only in the 44%, is this because there is something missing in the code? #8

Open blue-q opened 1 month ago

damNull commented 1 month ago

The code repo is complete and reproducible. The inconsistent performance may due to data processing error. You can refer to issue #6 and #7.

blue-q commented 1 month ago

Hi @damNull, thanks for your reply. How much "loop_times" should be set in the training process?

damNull commented 1 month ago

The experiments about loop_times are in Table 5 of Arxiv article.

blue-q commented 1 month ago

@damNull yeah, I now set loop_timesto 0,1,2 respectively for training, both on ntu60 and 120 with only 40 + accuracy. I only modified the value of loop_timesin the model, and I also processed the data according to readme.md. Could you please help me analyze the reason?