Test accuracy is low than original test accuracy in the paper.

[redhat12345]

Test accuracy is low than original test accuracy in the paper. Can you fix it please?

[SSARCandy]

That's true, i will try to improve it

[redhat12345]

In original implementation, they used pre-trained model. Did you use pretrained model ?

[SSARCandy]

I did use ImageNet pre-trained model,
please see https://github.com/SSARCandy/DeepCORAL/blob/master/main.py#L147

[redhat12345]

Yes.. Sorry for the comment without noticing. Thank you for your kind reply. I did not find the reason of low accuracy?

[SSARCandy]

Although the accuracy is lower than paper's claimed, I think the difference (improvement) between w/ CORAL and w/o CORAL is still noticeable.

But the low accuracy is still a problem, I'll try to find out why.

[redhat12345]

Yes...absolutely.

[SSARCandy]

I think we can narrow down the accuracy issue to the w/o CORAL loss test task, since even w/o CORAL loss, the accuracy on test case didn't as good as paper claimed. (my: ~47%, paper: ~55%)

I didn't find out why😞, @redhat12345 could you help with this issue..? thanks!

[ZhangJUJU]

these days I am tryingto find the problems,but I still not fix it yet.I will try to fix it.not I am trying to write an custom dataloader,I guess the problem caused by the pytorch dataloader.plz wait a few days..if I can fix it.

[ZhangJUJU]

Sorry ,I made a mistake.Today I try another method to shuffle the dateset,but failed...I'll continue try to fix the problems.

[redhat12345]

@SSARCandy
I think the problems comes from data preprocessing. The way pytorch process data is different than caffe. I am also trying to fix the problems.

[ZhangJUJU]

@redhat12345 you may rewrite the dataloader or otherwise....

[redhat12345]

@SSARCandy

I will writedown dataloder script and update you.

[redhat12345]

Did you try in the following way?
(http://pytorch.org/docs/master/notes/autograd.html)

model = torchvision.models.alexnet(pretrained=True)
for param in model.parameters():
param.requires_grad = False
model.fc = nn.Linear(4096, 31)

[ZhangJUJU]

@redhat12345 @SSARCandy In my view,Change the net structure from Alexnet to Resnet18 or other,is not a good way to solve the low accuracy,becauce the algrithm yours compaire with the mainstream algorithms always use Alexnet as the net struct,in paper,the author compaire his algritm with the oters using Alexnet and GoogLeNet,all the algrithm in GoogleNet has better performance than Alexnet.So we should continue to find the problem in pytorch's training step or dataloader.