llava_qwen_slowfast_w2s_stage3

Generalizable Video Quality Assessment via Weak-to-Strong Learning

This repository contains the model weights for the paper Generalizable Video Quality Assessment via Weak-to-Strong Learning.

Video quality assessment (VQA) seeks to predict the perceptual quality of a video in alignment with human visual perception. This work explores weak-to-strong (W2S) learning as a new paradigm for advancing VQA without reliance on human-labeled datasets, achieving state-of-the-art results across both in-domain and out-of-distribution benchmarks.

• GitHub Repository: https://github.com/clh124/W2S-VQA

Installation

pip install transformers==4.44.0 pytorchvideo

Usage

To use this model, follow the feature extraction and inference steps provided in the official repository.

1. Extract SlowFast features

Navigate to the slowfast_feature directory and run:

python extract_slowfast_feature.py --feature_save_folder path/to/save_features --videos_dir path/to/videos

2. Run Inference

Ensure your test JSON file follows the format specified in the GitHub README, then run:

python infer_pair.py

Citation

@article{cao2025breaking,
  title={Breaking annotation barriers: Generalized video quality assessment via ranking-based self-supervision},
  author={Cao, Linhan and Sun, Wei and Zhang, Kaiwei and Peng, Yicong and Zhai, Guangtao and Min, Xiongkuo},
  journal={arXiv e-prints},
  pages={arXiv--2505},
  year={2025}
}

@article{cao2025towards,
  title={Towards Generalized Video Quality Assessment: A Weak-to-Strong Learning Paradigm},
  author={Cao, Linhan and Sun, Wei and Zhu, Xiangyang and Zhang, Kaiwei and Jia, Jun and Peng, Yicong and Zhu, Dandan and Zhai, Guangtao and Min, Xiongkuo},
  journal={arXiv preprint arXiv:2505.03631},
  year={2025}
}