[WIP][DiffusionDet] Diffusion models for object detection

[HichTala]

This pull request introduces a new model, DiffusionDet, and its associated components. The changes include the addition of configuration, head, loss, and model files, as well as a temporary debug script to test the model.
This is a first draft, nothing ready to use right now, I am creating this PR to get some feedback as it is my first time.

Key changes:

New Model Addition:

• src/diffusers/models/diffusiondet/modeling_diffusiondet.py: Added the DiffusionDet class to implement the DiffusionDet model. This includes the model's initialization and forward pass.

Configuration:

• src/diffusers/models/diffusiondet/configuration_diffusiondet.py: Added the DiffusionDetConfig class to handle the configuration settings for the DiffusionDet model.

Head Implementation:

• src/diffusers/models/diffusiondet/head.py: Added the DiffusionDetHead class and supporting functions for the model's head, including the ROIPooler class for region of interest pooling.

Loss Computation:

• src/diffusers/models/diffusiondet/loss.py: Added the DynamicCriterion class to compute the loss for the DiffusionDet model, including the process for Hungarian assignment and supervision of matched pairs.

Debug Script:

• debug_diffusiondet.py: Added a script to create and test the DiffusionDet model using the configuration and model classes.# What does this PR do?

Fixes # (issue)

• DiffusionDet: Diffusion models for object detection #1350

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Did you read our philosophy doc (important for complex PRs)?
• Was this discussed/approved via a GitHub issue or the forum? Please add a link to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@patrickvonplaten and @ShoufaChen participated in the previous implementation trial.

[HichTala]

The original model uses lists of detectron Bbox objects, I am adapting the code to use classical pytorch tensor with shape (B, N, 4) (B: batch size, N: number of proposals)

[yiyixuxu]

hi @HichTala thanks for the PR, the model is a bit of out dated at this point I think
do you want to try to see if we can make it work with remote_code? https://huggingface.co/docs/diffusers/using-diffusers/custom_pipeline_overview#community-components

[HichTala]

Hi @yiyixuxu,
Thank you for the suggestion! I’m happy to explore making it work with remote_code. My main goal is to contribute to the community, and Hugging Face is a great platform to share this work. I’m open to following whatever approach the main contributors believe is the best way to share it. Let me know how we can proceed!