samin9796 / ColorFoil

ColorFoil: How Well Can Vision and Language Models Perceive Colors?

Abstract: To be added. (Under construction)

Instructions

1. Link to download the MS COCO validation set (2017): https://cocodataset.org/#download
2. Run the zero_shot_evaluation.ipynb file. It will construct the ColorFoil benchmark from the MS COCO dataset and perform the zero-shot evaluation of CLIP, ViLT, and BridgeTower models.