Testing too slow

Open AymanMIbrahim opened this issue 5 years ago • 2 comments

Hello sir, great job really it's amazing but the testing on 91.pth is really slow it takes around 37 sec per image using google colab any help pease??

Jun 30 '20 20:06 AymanMIbrahim

Exactly this line of code

with torch.no_grad(): 
      y, feature = net(x)

takes all the time in the world in the new model is there any solution

Jun 30 '20 23:06 AymanMIbrahim

same issue encountered. I found that it is due to the denormal numbers (< 1E-32) in the weights, please refer to https://discuss.pytorch.org/t/conv2d-is-very-slow-on-trained-weights-vs-random-weights/43377/4 BTW, which dataset did you use for training?

Dec 22 '21 02:12 yayagege