关于进行test指令运行遇到错误

[king-huoye]

我在运行该指令：python Dehazing/test.py --size 'L' --input_dir '/data/Dense-Haze/Test/Haze' --target_dir '/data/Dense-Haze/Test/GT' --result_dir 'pre_process_Dense' --weights 'Dehazing/pretrai
ned_models/densehaze-MB-TaylorFormer-L.pth'

出现了这样的结果：===>Testing using weights: Dehazing/pretrained_models/densehaze-MB-TaylorFormer-L.pth
C:\ProgramData\Anaconda3\envs\taylor\lib\site-packages\torch\cuda_init_.py:104: UserWarning:
NVIDIA GeForce RTX 4090 with CUDA capability sm_89 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_61 sm_70 sm_75 compute_37.
If you want to use the NVIDIA GeForce RTX 4090 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

warnings.warn(incompatible_device_warn.format(device_name, capability, " ".join(arch_list), device_name))
0it [00:00, ?it/s]
C:\ProgramData\Anaconda3\envs\taylor\lib\site-packages\numpy\core\fromnumeric.py:3441: RuntimeWarning: Mean of empty slice.
out=out, **kwargs)
C:\ProgramData\Anaconda3\envs\taylor\lib\site-packages\numpy\core_methods.py:189: RuntimeWarning: invalid value encountered in double_scalars
ret = ret.dtype.type(ret / rcount)
final PSNR: nan final SSIM: nan
请问该如何解决呢

[snowy-land]

您好，请问您现在解决了吗？我也遇到了同样的问题

[raindrop313]

@king-huoye @snowy-land 可能是由于您所使用的pytoch版本较旧，据我所知4090不支持一些老版本的pytorch，建议安装新的pytorch版本，或是使用老的显卡。
可以参考下这个帖子：‘discuss.pytorch.org/t/pytorch-not-working-with-nvidia-4090/173054‘

[snowy-land]

@raindrop313 谢谢您，我在更换到3090并更改环境版本后解决了第一个问题算力匹配的问题。但是我运行了指令：python Dehazing/test.py --size 'L' --input_dir '/data/Dense-Haze/Test/Haze' --target_dir '/data/Dense-Haze/Test/GT' --result_dir '/data/Dense-Haze/pre_process_Dense' --weights 'Dehazing/pretrained_models/densehaze-MB-TaylorFormer-L.pth'

之后出现了这样的结果：
===>Testing using weights: Dehazing/pretrained_models/densehaze-MB-TaylorFormer-L.pth
0it [00:00, ?it/s]
/root/miniconda3/envs/pytorch181/lib/python3.7/site-packages/numpy/core/fromnumeric.py:3441: RuntimeWarning: Mean of empty slice.
out=out, **kwargs)
/root/miniconda3/envs/pytorch181/lib/python3.7/site-packages/numpy/core/_methods.py:189: RuntimeWarning: invalid value encountered in double_scalars
ret = ret.dtype.type(ret / rcount)
final PSNR: nan final SSIM: nan
请问这要怎么解决呢

[raindrop313]

因为SOTS数据集和densehaze数据集命名不太相同，如果你是想在densehaze上面推理的话，可以尝试修改test.py中的这个代码：
’target=np.float32(utils.load_img(os.path.join(target_dir,file_.split('/')[-1])))/255.‘
改成如下形式：
’target=np.float32(utils.load_img(os.path.join(target_dir,file_.split('/')[-1].split('_')[0]+'_GT.png')))/255.‘
可以尝试下能否解决你的问题
@snowy-land

[king-huoye]

@raindrop313 不过我装的pytorch是官网最新的那个118的那个，我最近在忙别的，我晚一点根据你的意见试一下哈

[snowy-land]

@raindrop313 十分感谢您抽出时间回复我。我还想请教下我在用ITS数据集训练时运行指令“sh train.sh Dehazing/Options/MB-TaylorFormer-L.yml”后，出现了这样的结果：
Traceback (most recent call last):
File "basicsr/train.py", line 325, in
main()
File "basicsr/train.py", line 186, in main
model = create_model(opt)
File "/root/autodl-tmp/project/Restormer/basicsr/models/init.py", line 38, in create_model
model = model_cls(opt)
File "/root/autodl-tmp/project/Restormer/basicsr/models/image_restoration_model.py", line 65, in init
self.net_g = define_network(deepcopy(opt['network_g']))
File "/root/autodl-tmp/project/Restormer/basicsr/models/archs/init.py", line 45, in define_network
net = dynamic_instantiation(_arch_modules, network_type, opt)
File "/root/autodl-tmp/project/Restormer/basicsr/models/archs/init.py", line 39, in dynamic_instantiation
raise ValueError(f'{cls_type} is not found.')
ValueError: MB_TaylorFormer is not found.

之后我将train.sh的pretrain_network_g修改到了ITS-MB-TaylorFormer-L.pth的路径，但还是有这个报错

[raindrop313]

你的环境的根路径是不是放在restormer文件夹下面了？打开MB_TaylorFormer的文件夹，然后：
“
python setup.py develop --no_cuda_ext
”
试下能否解决。
这个问题不是权重路径不对引起的，二十basicsr在初始化时没有找到MB_TaylorFormer.py的路径
@snowy-land

[snowy-land]

@raindrop313 非常感谢您这么晚还回复我，在执行您所说的操作后我的问题解决了，谢谢

[snowy-land]

@raindrop313 您好，我想再次请教下要如何在训练时输出SSIM呢