search

alibaba / conv-llava

Apache License 2.0
101 stars 3 forks source link

Evaluation on RefCOCO #5

Closed jungle-gym-ac closed 3 months ago

jungle-gym-ac commented 3 months ago

Hello, I want to reproduce the results on RefCOCO, RefCOCO+ and RefCOCOg, but I found that VLMEvalKit doesn't support these datasets, and lmms-eval only support REG(Grounded Captioning) evaluation on these datsets. Did you evaluate REC(Visual Grounding) Evaluation results with your own scripts? Thank you!

John-Ge commented 3 months ago

I have update the evaluation code: https://github.com/alibaba/conv-llava/blob/main/scripts/refcoco.sh You could directly use above code for evaluation.