Darknet: Resnet Throws Error

[craterkamath]

I'm trying to train ResNet152_YOLO on my custom data.

I have downloaded a CFG from this link:
https://drive.google.com/drive/folders/1rof0f8mCNKTYJOCWElTYQLOq513Gh5_V

I have altered the Input width and height to 256.
I have 4 classes: I've changed the [convolutional] and [region] section also.

The program shows corresponding error. I tried using another cfg file also, It didnt change the output.

[craterkamath]

https://github.com/pjreddie/darknet/issues/187 Talks about how the darknet has resizing issues, I thought that was resolved in a later update.

[AlexeyAB]

New [shortcut] layer supports multiinput, weights and normalization, but doesn't support different width & height of input layers.

I will think how can I fix it.

[craterkamath]

Update: I tried running the same cfg using the native darknet repo pjreddie/darknet, it works just fine.

[AlexeyAB]

Yes, because new [shortcut] layer supports multiinput, weights and normalization, but doesn't support different width & height of input layers.

[craterkamath]

@AlexeyAB Any updates on this issue?

[zunairaR]

I am having the same problem @craterkamath and @AlexeyAB . Can anyone help???