Open iamhankai opened 4 years ago
之前的实验对第一个卷积层进行了修改,最近也补充了未修改第一个卷积层的实验。两个实验分别在 /results/ghost-vgg-16-not-1st-conv /results/ghost-resnet-56-not-1st-conv
里面有log,但是acc仍旧未有改善,您有时间的话,帮忙看看log里有哪些参数设置不正确的么,非常感谢!
There are duplicated BN and ReLU in your code. The learning rate only decay once which is not enough.
Isparallel = True if len(args.gpu) > 1 else False
if len(args.gpu) > 1:
Isparallel = True
else:
Isparallel = False
gpu设置的时候出现这种问题这个怎么解决
Traceback (most recent call last):
File "D:/rhnet-daima/ghostnet_cifar10-master/bin/02_main.py", line 47, in
Isparallel = True if len(args.gpu) > 1 else False if len(args.gpu) > 1: Isparallel = True else: Isparallel = False gpu设置的时候出现这种问题这个怎么解决 Traceback (most recent call last): File "D:/rhnet-daima/ghostnet_cifar10-master/bin/02_main.py", line 47, in Isparallel = True if len(args.gpu) > 1 else False TypeError: object of type 'NoneType' has no len()
看起来你没有设置gpu id?
Have you modified the 1st conv layer to Ghost module? We did not touch the 1st conv in our experiments.
https://github.com/iamhankai/ghostnet.pytorch/issues/16