jin-ying/gta-seg

Code for GTA-Seg (NeurIPS2022)

gta-seg's Issues

Data spilt for Pascal VOC12.

Does the data split of Pascal VOC12 in GTA-Seg is same as U2PL ? And if you follow the U2PL ,all the experiments are 100% high-quality labels within the AUG experiments, including 2646, 1323, 732, 366, maybe it is unfair. Could you please report results of the spilt same as CPS？
Thanks！

When will you be able to make the source code public?

I appreciate your work very much, when will you be able to make the source code public?

关于论文中公式6的理解

您好，感谢您卓越的工作，我不太能理解公式6中的每个像素权重的意义，i和j应该表示这个图片中的像素位置，那为什么还要在公式中加入两个对i和对j的求和符号。希望得到您的回复。

A question about batchsize

Hi Ying, thank you for your codes and the great work! I have one question about the batch size.

In the config, the batch size is 3 per GPU. So for each GPU (4 in total), it will have 3 labeled, 3 unlabeled for the teacher model, and 3 unlabeled for the GTA model.

Do I understand right? Appreciate your reply.