FAIR: Fine-grained Alignment and Interaction Refinement

This repository contains the official PyTorch implementation of our paper Towards Fine-Grained Adaptation of CLIP via a Self-Trained Alignment Score.

---

📁 Dataset Setup

• Dataset paths are defined in json_files/dataset_catalog.json.
  You will need to update these paths to point to your local dataset locations.

• Dataset label files are provided in json_files/classes.json.

• For dataset preparation, we recommend using the scripts from the VISSL repository.

---

📦 Requirements

Make sure to install the following dependencies:

• torch >= 1.10.0
• timm == 0.4.12
• tensorboardX
• ftfy

---

🚀 Training

To train the FAIR model, use the following command:

python train.py --dataset [name_of_dataset] --train_config ours_base --reg_batch

---

🙏 Acknowledgements

This work builds upon the codebase of MUST.
We thank the authors for publicly releasing their implementation.

If you use FAIR or find it helpful, please consider citing our paper.