search

siyuanliii / masa

Official Implementation of CVPR24 highligt paper: Matching Anything by Segmenting Anything
https://matchinganything.github.io
Apache License 2.0
784 stars 40 forks source link

Prompt Engineering Techniques #13

Open jackdong8588 opened 1 week ago

jackdong8588 commented 1 week ago

Can you tell me what kind of prompt words should generally be filled in after the parameter --texts to accurately recognize the object? Directly filling in the name of the female lead in the movie would definitely not be recognized. Besides "girl" and "woman," can appearance descriptions be used for detection?

jackdong8588 commented 1 week ago

"woman in red dress" "man with glasses" "girl with blonde hair" "yellow car" "blue bicycle"

Can the Detic_new model adapt to the above prompt words?

siyuanliii commented 1 week ago

Detic is not very good at those grounding expressions. Grounding DINO might be a better choice and yes, we can support those expressions, you can try!