Add MM Grounding DINO

[rziga]

Model description

Hi,

MM Grounding DINO is MMDetection's implementation of Grounding DINO. It improves zero-shot detection performance quite a bit (especially on LVIS) and is very similar to the original Grounding DINO implementation-wise. I'm mostly interested in it because they also have checkpoints for models with higher resolution FPN features, which are not available for the original model.

The model also appears to be the same inference-wise as LLMDet (#37334), so this could be a good stepping stone for that one.

Open source status

• The model implementation is available
• The model weights are available

Provide useful links for the implementation

Weights and implementation are available in MMDetection here.

I've already started working on porting the model with modular transformers, so I can open a draft PR if you want.

[Rocketknight1]

cc @qubvel @NielsRogge

[demoncoder-crypto]

Is this open for Community contribution?

[qubvel]

Hi @rziga, that sounds super nice! Feel free to ping me when PR is open!

@demoncoder-crypto thanks for your interest, you can collaborate with @rziga if they need any help 🤗