Add in_chans to DetrModel

[Zhylkaaa]

🚀 Feature request

Add in_chans parameter to DetrConfig and corresponding models

Motivation

I find myself wanting to test DETR on data that have only one channel (it can be grey scale images or spectrograms in my case). But to make it work I have to use hacks, that may eventually work purely.

Your contribution

I will be happy to open PR with this if this is as useful as I see it. Maybe some other improvements that can be incorporated?
UPADTE: I think this will involve also adding parameters to disable replace_batch_norm and very specific part of setting parameter.requires_grad_(False) to all parameters in resnet except layer2-4

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[LunaJang]

I also think it could be useful. I don't want to waste memorys by convert 1 channel image to 3 channels.