[rocauc]

A suggestion: I'd swap llava for Florence-2 for your open set text description. Florence-2 seems uniformly more descriptive in its outputs.

[Eisenstein]

They are using Ollama which is based on llama.cpp; florence is not supported on that backend.

[jerpint]

I found grounding-dino better than Florence and faster

[netdur]

I found YOLOS to be faster and better, bot real time but 22k objects under half second