/CVLM

Primary LanguagePythonMIT LicenseMIT

CVLM

CVLM: A Multimodal VLLM

Framework

framework.png

Evaluate

MME

model_path=ToOverwrite # trained model to replace  
image_path=MME_Path # MME testset path  
bash scripts/evaluation_mme.sh $model_path $image_path  

2023-11-18: CVLM has achieved 1636.46 perception score (No.1), 448.93 cognition score(No.2), and 2125.39 points in total (No.1). Please refer to MME

Acknowledgement

CLIP
MiniGPT-4
LLaVA
Vicuna

Copyright © 2023 CVLM
CVLM team members: apolloluo gonglujin buptlihang yjhdhr zrczrczrc uazx000 zonefv AndyDu0116 mineh810 etc.