Kingking
Kingking
when I run the generate_mask.py,I found that UNet11 and UNet16 can not be used. I don't know why.Anybody can help me ?
  我下载了您提供的代码，没有做出任何改动，用单卡P100 16G在CUB数据集训练 测试结果却达不到您在论文上报告的精度，比如在top1 loc acc上大概相差一个点 请问，如何能够达到与论文相近的精度 谢谢
The final output of my segmentation network is a single-channel prediction map, how can this situation be adjusted according to the tutorial?
about BN
Why not use BN?
我发现PFENet.py 中的 210--supp_feat_4 = self.layer4(supp_feat_3*mask) 将supp_feat_3 与 mask结合作为 layer4的输入，这一步的设计好像没有在论文中提及。 请问这一步关键吗
 作者您好，在1-shot的setting下，您是： 1.训练完一个split后，接着训练另外一个split，直到4个split训练完； 还是： 2. 4个split一起训练，每个split各占一张GPU； 我觉得按照第一种方式，训完4个split需要的很长时间， 但我尝试第二种方法，4个split一起训练，每个split占用一张GPU，4个程序都启动后训练速度都会下降 请问您有没有更加高效的训练策略
https://github.com/juhongm999/hsnet/blob/6b1bc12b3ff9f3950f49c55ab6f089e8d9e0d191/train.py#L19 I found an issue where the line19 code would re-fix the random seed with the same parameters every time after the second epoch, which would make the sampled pairs...
The same model is trained multiple times, and the process and final result of each training are different. How can I use your code to ensure stable reproduction?
I can get 80.x mIoU in the validation set with the weights you provided. But I trained myself from scratch, I could only get 40.x mIoU.
 关于SegGPT在FSS的实验中有点困惑，我想知道SegGPT在训练的时候是否已经用上Pascal和COCO 的所有类别？ 据我的了解，过去FSS的实验中，训练集的类别与测试集的类别是不交叉的。 而论文中说：For a fair comparison, we also evaluate specialist models on in-domain categories marked by *. * indicates that the categories in training cover the categories in...