Hi, I note that you benchmark the computation of STDC2-Seg75 which is not reported in the CVPR2021 paper. Did you test the speed of STDC-Seg on your own platform? How about the results?

I try show.py. But I can not.

$ python3 show.py
name= cityscapes
train size: 2975
val size: 500
Traceback (most recent call last):
  File "show.py", line 358, in <module>
    show_cityscapes_model()
  File "show.py", line 337, in show_cityscapes_model
    show(model,val_loader,device,show_cityscapes_mask,num_images=num_images,skip=skip,images_per_line=images_per_line)
  File "show.py", line 134, in show
    outputs = model(images)
  File "/home/sounansu/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/sounansu/RegSeg/model.py", line 76, in forward
    x=self.stem(x)
  File "/home/sounansu/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/sounansu/RegSeg/blocks.py", line 22, in forward
    x = self.conv(x)
  File "/home/sounansu/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/sounansu/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 446, in forward
    return self._conv_forward(input, self.weight, self.bias)
  File "/home/sounansu/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 442, in _conv_forward
    return F.conv2d(input, weight, bias, self.stride,
RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

Hi, thank you for sharing the code. Can you provide download link about the pretrained model(exp48_decoder26 and exp53_decoder29) in Cityscapes dataset, Thank you very much!

When using seg_transforms.py through your scripts 'camvid_efficientnet_b1_hyperseg-s', there always exsist 'TypeError: resize() got an unexpected keyword argument 'interpolation'' in 174 line. Does this bug only appear in this scripts and should I modify the code when using this scripts?

Patching CVE-2007-4559

Hi, we are security researchers from the Advanced Research Center at Trellix. We have began a campaign to patch a widespread bug named CVE-2007-4559. CVE-2007-4559 is a 15 year old bug in the Python tarfile package. By using extract() or extractall() on a tarfile object without sanitizing input, a maliciously crafted .tar file could perform a directory path traversal attack. We found at least one unsantized extractall() in your codebase and are providing a patch for you via pull request. The patch essentially checks to see if all tarfile members will be extracted safely and throws an exception otherwise. We encourage you to use this patch or your own solution to secure against CVE-2007-4559. Further technical information about the vulnerability can be found in this blog.

If you have further questions you may contact us through this projects lead researcher Kasimir Schulz.

When training, how did the miou and accuracy calculate? On train dataset or validate dataset? I think it's calculated on val dataset due to https://github.com/RolandGao/RegSeg/blob/main/train.py#L238. I trained the base regseg model with config cityscapes_trainval_1000epochs.yam on Cityscapes and got the unbelievable results.

Hi, RolandGao, nice to see a good job! I see you've done a lot of experiments on the backbone setting, but I still have some confusion after reading your published paper.

• First, You calculate the fov of 4095 to see the bottom-right pixel when training cityscape (1024x2048), so you have verify the backbone should be exp48 [ (1,1) + (1,2) + 4 * (1, 4) + 7 *(1, 14) ] with fov (3807). But I also find the same backbone when training the CamVid (720x960). Why not use a shallow backbone? I am training my own dataset with image resolution (512 x 512), do I need to modify the backbone architecture? Can you give some advice?
• Second, I test inference time of regseg. I notice that the speed is not better than other real-time archs due to split and dilated conv even if model costs low GFLOPs. In the application, what we are concerned about is the speed, so is there any strategy to improve the speed?

Hi, Thanks for your excellent work ! I notice that RegSeg can achieve a high accuracy on Cityscapes without pretraining. I also did a lot of ablation studies and I think DDRNet will drop around 3% miou if they do not use ImageNet pretraining. How about trying to train your encoder on ImageNet and see what will happen? I really look forward to your result ! Thanks !