images in kitti dataset cannot match

[Shelfcol]

I want to implement the algorithm in kitti dataset. But it shows that the codes cannot match kitti dataset's images' size.

CUDA_VISIBLE_DEVICES=0 python3 demo.py --ckpt res/model_final.pth --img_path ./2.png
Traceback (most recent call last):
File "demo.py", line 39, in
im = to_tensor(Image.open(args.img_path)).unsqueeze(0).cuda()
File "/home/gxf/anaconda3/envs/tf1.4.1/lib/python3.6/site-packages/torchvision/transforms/transforms.py", line 61, in call
img = t(img)
File "/home/gxf/anaconda3/envs/tf1.4.1/lib/python3.6/site-packages/torchvision/transforms/transforms.py", line 166, in call
return F.normalize(tensor, self.mean, self.std, self.inplace)
File "/home/gxf/anaconda3/envs/tf1.4.1/lib/python3.6/site-packages/torchvision/transforms/functional.py", line 217, in normalize
tensor.sub_(mean[:, None, None]).div_(std[:, None, None])
RuntimeError: output with shape [1, 370, 1226] doesn't match the broadcast shape [3, 370, 1226]

I dont know if you can spare some time to fix this problem. Thank you.

[CoinCheung]

The reason is likely to be that your image is an one-channel image, likely to be gray. I have added a bit fix, you can try again.

[Shelfcol]

It worked.thx
By the way, last question you suggested me to chang the code in demo.py from fp16.model import BiSeNet to from model import BiSeNet. Maybe you can change your code in demo.py in your repo, and prevent others from meeting the same error .