returned label differs from original prompt in Grounded-SAM2

[mtablo]

Issue

Hi, I'm attempting to detect "a black and white french bulldog" from the image using Grounded-SAM2.

While the model successfully detects the dog mask, but the results[0]["labels"] from the post_process_grounded_object_detection function is "a black french bulldog", which differs from the original prompt.

Question

I want the exact text that was in the original prompt.
How can I ensure that the detected label exactly matches the original prompt text?

[rentainhe]

Hi @mtablo , we have some similar discussions here: #50 , you can try the local grounding dino demo and follow our Grounding DINO specific phrase inference instructions to achieve this.

[mtablo]

thank you! I'll try that!