You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hey, thanks for your work.
Have you tested the accuracy for Cifar10?
I tried to quantized ResNet20 into 2/2 4/4 bit, and the accuracy only 88%~90%.
If possible, can you share you result?
The text was updated successfully, but these errors were encountered:
Please read the existing issues. We have discussed it yet.
Some hyper parameters in the paper only work for ImageNet. So you need to adjust them manually for CIFAR10.
The LSQ is not very robust to the initial value. You can read "LSQ+" for more details.
Hey, thanks for your work. Have you tested the accuracy for Cifar10? I tried to quantized ResNet20 into 2/2 4/4 bit, and the accuracy only 88%~90%. If possible, can you share you result?
Hello I am having trouble with training with the same setting, as I am getting abnormally high result. I would be grateful if you could help me figure out what might be going wrong.
Hey, thanks for your work.
Have you tested the accuracy for Cifar10?
I tried to quantized ResNet20 into 2/2 4/4 bit, and the accuracy only 88%~90%.
If possible, can you share you result?
The text was updated successfully, but these errors were encountered: