Yang Zou
You can keep the batch size as 2 while reduce the input image size. You can try to reduce the '--input-size' to smaller size (480, 960) Check https://github.com/yzou2/CRST/issues/2#issuecomment-549638137
Maybe you don't have enough hard disk space. On Thu, Jul 2, 2020 at 2:24 AM xuebingbing121 wrote: > OK thanhs, I will try, But the first round of training...
You can keep on trying until it works.
The results can be seen in the *round_name/ folder. The output probability map will be generated on the fly and saved in *round_name/prob. But since it will consume lots of...
The probability maps should be saved in the /prob folder. In addition, dropout can be added deeplab/model.py
You can try to add dropout but you may need to do some network surgery to use the pretrained deeplab weights. But I think it's better to retrain the network...
Your questions are quite general. You can easily find tutorials of deep learning online. The github issues are used to answer the questions of the specific repository and code. I...
It looks like it's caused by the failure in model loading. I don't get the bug in my computer. Alternatively you may try to download the resnet101 to local file...
I borrowed the deeplab training code from https://github.com/speedinghzl/Pytorch-Deeplab . The ResNet_Refine is set there but not tested for CRST/CBST. If you want to use it, you may need to modify...
The results are generated based on DeepLabV2. If you want to use ResNet-38, you can add confidence regularization into our CBST repo which utilizes ResNet-38: https://github.com/yzou2/CBST