Error when running iw3——“PytorchStreamReader failed reading zip archive: failed finding central directory”

[yiyan32]

image
Strictly follow the steps to Manually install（https://github.com/nagadomi/nunif/blob/master/INSTALL-windows.md. The waifu2x is running well.
However, when using iw3, whether it is converting pictures or videos, the same error will occur. The parameter settings are as shown above.

[nagadomi]

If waifu2x works, the iw3 model file may be corrupted due to download problems.
Try deleting iw3/pretrained_models/hub folder and try again.
The first run will download a 1GB+ file, so it may take a long time.

Also, CLI will provide more detailed error messages.

python -m iw3.cli -i input.jpg -o output.png

[yiyan32]

I have downloaded all of the 3 depth models, still reporting an error. using the code you provided above for testing, the error details are as follows

thank you for your time

[nagadomi]

The error is iw3/pretrained_models/hub/checkpoints/ZoeD_M12_N.pt file loading error.
Pytorch checkpoint file(model/weight/parameter file, .pt,.pth) is internally a zip archive, so the file format error seems to be such a message.

Possible causes are,

1. File is corrupt
2. File path problems (e.g., multibyte characters, read permissions)
3. Python or PyTorch version compatibility (pickle version)

It's not likely to be 2 or 3, as waifu2x(loading other checkpoint files) is working.
Possible causes of file corruption,

• Download Failure
• Disk full
• Modification by anti-virus software, etc.

Could you check the size and MD5 of iw3\pretrained_models\hub\checkpoints\ZoeD_M12_N.pt?

1. check file size (on command prompt)

dir iw3\pretrained_models\hub\checkpoints

result

 Volume in drive C has no label.
 Volume Serial Number is EA34-A50B

 Directory of C:\dev\nunif\windows_package\nunif\iw3\pretrained_models\hub\checkpoints

09/27/2023  07:43 AM    <DIR>          .
09/27/2023  07:43 AM    <DIR>          ..
09/27/2023  07:43 AM     1,443,406,099 ZoeD_M12_N.pt

2. check md5sum

certutil -hashfile iw3\pretrained_models\hub\checkpoints\ZoeD_M12_N.pt MD5

result

MD5 hash of iw3\pretrained_models\hub\checkpoints\ZoeD_M12_N.pt:
f9b240db5a30a1600e25cad6866516b7
CertUtil: -hashfile command completed successfully.

File size: 1,443,406,099
MD5: f9b240db5a30a1600e25cad6866516b7

same result on linux.

% ls -la iw3/pretrained_models/hub/checkpoints/ZoeD_M12_N.pt
-rw------- 1 nagadomi nagadomi 1443406099  7月 28 13:02 iw3/pretrained_models/hub/checkpoints/ZoeD_M12_N.pt
% md5sum iw3/pretrained_models/hub/checkpoints/ZoeD_M12_N.pt
f9b240db5a30a1600e25cad6866516b7  iw3/pretrained_models/hub/checkpoints/ZoeD_M12_N.pt

Those ZoeDepth model files are downloaded from https://github.com/isl-org/ZoeDepth/releases/tag/v1.0

[yiyan32]

It is indeed an error caused by the damaged model. After manually downloading the corresponding model from Huggingface, it can run smoothly. Thank you for your patience and guidance!

[nagadomi]

Do you have any idea why the file is damaged?

[nagadomi]

It seems that torch.hub.load stores the files first to the system's temporary folder.
Possibly there is not enough space there.

I will add file validation with the hash value.

[wangchuanzhang]

Why is only the CPU recognized here and not my GPU? The CPU can be used, but it's too slow.

[yiyan32]

The above pic shows the model size and hash value that I downloaded automatically through the gui, while the system disk space is enough, it is very likely a network problem.

[nagadomi]

Thanks for the reply.
It seems that there are repeated download failures, so automatic download retries may not fix the problem.
However, I will try to clarify the error message.