Spaces:

---

gokaygokay

/

Florence-2

like 776

Running on Zero

is it possible for fal to add a better or more accurate florence-2 obj detect endpoint?
fal's current implementation, it seems that the image is placed within a large white plane (why?), which makes all the detected bounding box values off

here is one that has the bounding boxes relative to original image:

https://huggingface.co/spaces/SkalskiP/better-florence-2