search

yxgeee / FD-GAN

[NeurIPS-2018] FD-GAN: Pose-guided Feature Distilling GAN for Robust Person Re-identification.
https://yxgeee.github.io/projects/fdgan.html
281 stars 80 forks source link

Stage III test #14

Closed LuoPeng-CV closed 5 years ago

LuoPeng-CV commented 5 years ago

Excuse me,I got a problem when I tested the best_net_E.pth in stage III. I can test the First stage evaluation, but when it tried to test the second stage,it always gave this information that RuntimeError: CUDA error: out of memory, and after I tried test it using several gpus(CUDA_VISIBLE_DEVICE),it didn't work as well. I'll deeply appreciate it if you could give some suggestions.Thank you! image

yxgeee commented 5 years ago

try to use a smaller batch size for testing

LuoPeng-CV commented 5 years ago

Thanks for your reply,I've tried to set batch size as 128,64,32,24, but all get the same mistake,RuntimeError: CUDA error: out of memory,and if i set batch size as 20(or less than 20), I got another problem as follow: image

yxgeee commented 5 years ago

try to set the num of threads to 0 in the testing data loader

LuoPeng-CV commented 5 years ago

I've tried your way,but it still get the same problem that everytime it reach 5500/16483. image I will try other ways to solve this problem.Anyway,sincerely thanks for your help!

heerduo commented 5 years ago

Hello, I encounter the same problem. Have you solved it?

LuoPeng-CV commented 5 years ago

Hello, I encounter the same problem. Have you solved it?

I remember I solved it at last,but I didn't remember it in detail. It's probably about modifying inputs = Variable(inputs, volatile=True).cuda() as with torch.no_grad(): inputs = Variable(inputs).cuda(). Hope it will help you.

heerduo commented 4 years ago

Can you point it in detail? "with torch.no_grad(): inputs = Variable(inputs).cuda()" does not work. Thanks.