lijin118 / cgdm Goto Github PK

From this line, it looks like you also use data augmentation during test-time. Are the results in the paper with this augmentation?

Hi, thanks for sharing. I had a question about the codes.

Testing
Epoch: 1 Test set: Average loss: -0.090460, Accuracy: 16960/32676 (51.903538%)
aeroplane: [4707/19740] (23.844985%)
bicycle: [12253/12936] (94.720161%)
Traceback (most recent call last):
File "train_visda.py", line 328, in
train(args.epochs + 1)
File "train_visda.py", line 275, in train
test(ep + 1)
File "train_visda.py", line 320, in test
100. * classes_acc[i][0] / classes_acc[i][1]))
ZeroDivisionError: float division by zero

Please help me, thanks a lot.

Hi, I am not able to understand the role of the entropy loss you use in your code, which was not mentioned in the paper.

Is this important to reproduce the result? Why is it not mentioned in the paper?

I visited the official site https://www.imageclef.org/2014/adaptation#test , but it seems that there is only test data here. How can I get the whole dataset? Or can you share the data?

您好，我最近在研究您的工作，发现在domainnet数据集上，acc先很高，后来下降了很多，并且聚类后的结果下降，请问这是正常的么

您好，在train_visda.py等程序中存在Entropy_both()函数，但是在其他文件中没有找到该函数的定义。