How to detect any object using caption inputs ?

[LuletterSoul]

Hello, thank you for sharing excellent work. Currently, the model calculates the similarity between text tokens and image features, selecting the top1 as its class. If I input a single sentence as text information (similar to Grounding DINO), will the model still work correctly? If so, how should it be modified?

[wondervictor]

You can find it at: #315 (comment). I'll add the caption input in the demo considering many requests about caption input.