CaPL

Causality-guided Prompt Learning for Vision-language Models via Visual Granulation

Paper Link: arXiv

The prompt learner is designed and initialized following SHIP (Wang et al., ICCV 2023). Please refer to initialization.

The attribute disentanglement module is designed following BBDM (Li et al., CVPR 2023). Please refer to BBDM-based network.

The codes for constructing factual granules and counterfactual granules are in granule.py.

test.py is used for inference, and the learned textual features are in weights (which are used solely for inference without additional modules).