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

Why ensemble accuracies are unstable? #5

Closed JasOlean closed 5 years ago

JasOlean commented 5 years ago

When I ensemble joint and bone score for every epoch, their accuracy is varying. I mean although same epoch score, different accuracy result. Why happened like that and how to fix it?

Anirudh257 commented 5 years ago

Check https://github.com/lshiwjx/2s-AGCN/issues/10 if that helps.

JasOlean commented 5 years ago

I cannot understand Chinese words, could you explain me how to fix it? @Anirudh257

Anirudh257 commented 5 years ago

You can use Google translate :)

kenziyuliu commented 5 years ago

Make sure you are ensembling the right scores for the right dataset, and not include any randomness on the testset when producing the classification scores

RayLeijingsi commented 5 years ago

I cannot understand Chinese words, could you explain me how to fix it? @Anirudh257

in ensemble.py, you can find that label[0], r1[0] and r2[0] are different. they are must be same. so, 'label', 'r1' and 'r2' need to be sorted. i hope it helps you.

JasOlean commented 5 years ago

@RayLeijingsi Very thank you Ray, Now I fix that.