Florence-2 phrase Grounding

[Hasanmog]

Any Florence-2 phrase grounding training script ?

[SkalskiP]

Hi @Hasanmog 👋🏻 Thank you so much for your interest in maestro. Not yet, unfortunately. I don't have any good dataset that I could use as an example. Do you have any ideas on what data or for what use case you would like to fine-tune?

[Hasanmog]

Hello @SkalskiP 👋🏻.
I would suggest using datasets like RefCOCO, RefCOCO+, and RefCOCOg to adapt the model for general phrase grounding tasks. Once the model is adapted, parameter-efficient fine-tuning methods like LoRA can be employed for domain-specific tasks, which is a computationally feasible alternative to full fine-tuning.