COSMIC (CVPR 2025)

Fanding Huang^{1 *}  Jingyan Jiang^{2 *}  Qinting Jiang¹  Hebei Li³  Faisal Nadeem Khan^{1 †}  Zhi Wang^{1 †}

^*These authors contributed equally to this work; ^†Co-corresponding authors.

¹Shenzhen International Graduate School, Tsinghua University, China

²Shenzhen Technology University, China

³University of Science and Technology of China, China

---

This repository is the official PyTorch implementation of "COSMIC: Clique-Oriented Semantic Multi-space Integration for Robust CLIP Test-Time Adaptation".

Environment Setting

To run this project, you need to set up your environment as follows:

conda create -n COSMIC python=3.10 -y
conda activate COSMIC
pip install -r requirements.txt

Dataset

For both Out-of-Distribution and Cross-Domain benchmarks, please refer to this page.

Run COSMIC

To execute the COSMIC, navigate to the scripts directory.

OOD Benchmark

Run COSMIC on the OOD Benchmark:

bash ./scripts/cosmic_ood.sh

Cross-Domain Benchmark

Run COSMIC on the Cross-Domain Benchmark:

bash ./scripts/cosmic_cd.sh

BibTeX

@inproceedings{huang2025cosmic,
  title={COSMIC: Clique-Oriented Semantic Multi-space Integration for Robust CLIP Test-Time Adaptation},
  author={Huang, Fanding and Jiang, Jingyan and Jiang, Qinting and Li, Hebei and Khan, Faisal Nadeem and Wang, Zhi},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={9772--9781},
  year={2025}
}