Pytorch-yolov3: Varying the number of input channels

[stadtgiraffe]

Thank you for the great work!
I got it working for my greyscale images since they are expanded to a three dimensional RGB image. But I wondered if its possible to simply change the code so the pipeline takes only one channel and computes faster (or even multiple channels > 3 for detection of 3D objects in series of concatenated Images).
And if yes, where do I have to change the code?

Thanks for the help!

[sub2sub2]

how to solve it?

[atinesh-s]

For processing single channel below changes should be done in the code

1. Changechannels=1 in config/yolov3-custom.cfg
2. Update line 47 and 86 in utils/datasets.py with desired image conversion mode

img = transforms.ToTensor()(Image.open(img_path).convert('1'))

https://pillow.readthedocs.io/en/stable/handbook/concepts.html#concept-modes

[dipam7]

For point 2, the following change worked for me

img = transforms.ToTensor()(Image.open(img_path)