search

SainingZhang / DDAMFN

97 stars 17 forks source link

Cannot reproduce the result using the checkpoint on RAF-DB dataset. #17

Open windJRK opened 1 month ago

windJRK commented 1 month ago

Has anything gone wrong with my result? I can only get 91.04% using pytorch-1.8.0.

wrr0212 commented 1 month ago

Has anything gone wrong with my result? I can only get 91.04% using pytorch-1.8.0.

Excuse me, are you using the aligned dataset or the original dataset?

windJRK commented 1 month ago

Has anything gone wrong with my result? I can only get 91.04% using pytorch-1.8.0.

Excuse me, are you using the aligned dataset or the original dataset?

Did you get the similar result? I use the retinaface to align the images to 112x112. When I use mobilenet0.25 as backbone, the accuracy is only lower than 91%. Why is it? Is it the problem of my dataset?

wrr0212 commented 1 month ago

Has anything gone wrong with my result? I can only get 91.04% using pytorch-1.8.0.

Excuse me, are you using the aligned dataset or the original dataset?

Did you get the similar result? I use the retinaface to align the images to 112x112. When I use mobilenet0.25 as backbone, the accuracy is only lower than 91%. Why is it? Is it the problem of my dataset?

My accuracy is not as high as yours yet.

windJRK commented 1 month ago

Has anything gone wrong with my result? I can only get 91.04% using pytorch-1.8.0.

Excuse me, are you using the aligned dataset or the original dataset?

Haha, this is normal. When I test on different graphics cards, I get different results, even both are 3090 graphics card.