ðPaper Arxiv | ð€Huggingface Dataset
Official Repository for CVPR 2025 paper Critic-V: VLM Critics Help Catch VLM Errors in Multimodal Reasoning.
git clone https://github.com/kyrieLei/Critic-V.git
cd Critic-V
cd data_utils/inject
python data_utils/inject/error_inject.py
cd data_utils/detect
python data_utils/detect/error_detect.py
After running this, for instance, you need to change the following name in your json files
"num_detections": "4",
"detection_1": "The woman is not playing with a small dog.",
"detection_2": "There is no picnic set up in the background.",
"detection_3": "A rainbow does not appear to be forming in the sky.",
"detection_4": "The woman is not just preparing to fly the kite; she is already holding it.",
to
"model1_num_detections": "4",
"model1_detection_1": "The woman is not playing with a small dog.",
"model1_detection_2": "There is no picnic set up in the background.",
"model1_detection_3": "A rainbow does not appear to be forming in the sky.",
"model1_detection_4": "The woman is not just preparing to fly the kite; she is already holding it.",
cd data_utils/score_compute/score_compute.py
python data_utils/score_compute/score_compute.py
If you find our code or paper useful, please cite as:
@article{zhang2024critic,
title={Critic-V: VLM Critics Help Catch VLM Errors in Multimodal Reasoning},
author={Zhang, Di and Lei, Jingdi and Li, Junxian and Wang, Xunzhi and Liu, Yujie and Yang, Zonglin and Li, Jiatong and Wang, Weida and Yang, Suorong and Wu, Jianbo and others},
journal={arXiv preprint arXiv:2411.18203},
year={2024}
}
