search

kenziyuliu / DGNN-PyTorch

Unofficial PyTorch implementation of the CVPR'19 paper "Skeleton-Based Action Recognition with Directed Graph Neural Networks".
Other
235 stars 59 forks source link

请问一下 我跑出的准确率为什么特别低 才80左右？空间流 #16

Closed dangao250 closed 4 years ago

dangao250 commented 4 years ago

Excuse me, why is the accuracy rate of my running out so low? Only about 80? Space flow

dangao250 commented 4 years ago

可以给邮箱 请教几个问题吗

dangao250 commented 4 years ago

Can you give me an email?

ltw18 commented 4 years ago

Excuse me, why is the accuracy rate of my running out so low? Only about 80? Space flow

你好，你解决了吗？这个代码给出的准确率只有85%，原文中接近90%，不知道原因

kenziyuliu commented 4 years ago

Hi there, unfortunately, I was not able to reproduce the reported accuracy of the paper. Given that I do not have spare hardware capacity for this, I think a good direction to investigate is https://github.com/kenziyuliu/DGNN-PyTorch/blob/master/model/dgnn.py#L52 where it was assumed that the temporal conv layers are shared between node/edge features (i.e. try using separate temporal layers)