Skip to content

Latest commit

 

History

History
106 lines (79 loc) · 6.11 KB

README.md

File metadata and controls

106 lines (79 loc) · 6.11 KB
Raw

VideoScore

This is the official repo for our EMNLP 2024 paper "VideoScore: Building Automatic Metrics to Simulate Fine-grained Human Feedback for Video Generation".


News

[2024-11-28] Try on our new version VideoScore-v1.1, with better performance in "text-to-video alignment" subscore and the support for 48 frames in inference now!

[2024-08-05] We released the Wandb training cruves of VideoScore and VideoScore-anno-only to help reproduce the training results.

Introduction

VideoScore.mp4

🚀The recent years have witnessed great advances in video generation. However, the development of automatic video metrics is lagging significantly behind. None of the existing metric is able to provide reliable scores over generated videos. 🤔The main barrier is the lack of large-scale human-annotated dataset.

  • 🛢️VideoFeedback Dataset. In this paper, we release VideoFeedback, the first large-scale dataset containing human-provided multiaspect score over 37.6K synthesized videos from 11 existing video generative models.

  • 🏅VideoScore. We train VideoScore (initialized from Mantis) based on VideoFeedback to enable automatic video quality assessment. Experiments show that the Spearman correlation between VideoScore and humans can reach 77.1 on VideoFeedback-test, beating the prior best metrics by about 50 points. Further result on other held-out EvalCrafter, GenAI-Bench, and VBench show that VideoScore has consistently much higher correlation with human judges than other metrics.

  • 🫡Human Feedback for Video generative models. Due to these results, we believe VideoScore can serve as a great proxy for human raters to (1) rate different video models to track progress (2) simulate fine-grained human feedback in Reinforcement Learning with Human Feedback (RLHF) to improve current video generation models.

Installation

  • for inference
pip install -e .
  • for evaluation
pip install -e .[eval]
  • for training
git clone https://github.com/TIGER-AI-Lab/Mantis
cd Mantis
pip install -e .[train,eval]
pip install flash-attn --no-build-isolation
# then training scripts are in Mantis/train/scripts

Dataset

  • 🤗 VideoFeedback VideoFeedback contains a total of 37.6K text-to-video pairs from 11 popular video generative models, with some real-world videos as data augmentation. The videos are annotated by raters for five evaluation dimensions: Visual Quality, Temporal Consistency, Dynamic Degree, Text-to-Video Alignment and Factual Consistency, in 1-4 scoring scale.

  • 🤗 VideoScore-Bench We derive four test sets from VideoFeedback, EvalCrafter, GenAI-Bench and VBench respectively to curate VideoScore-Bench. VideoScore-Bench is composed of about 7,000 videos, covering both Likert-scale annotation and human preference data.

Model

Inference examples

cd examples
python run_videoscore.py

Evaluation

For details, please check benchmark/README.md

Training

For details, please check training/README.md

Acknowledgement

  • Thanks Mantis for the training codebase of VideoScore (and variants) and also for the plug-and-play MLLM tools in evaluation stage!

  • Thanks VIEScore for some codes of prompting MLLM in evaluation!

Citation

@article{he2024videoscore,
  title = {VideoScore: Building Automatic Metrics to Simulate Fine-grained Human Feedback for Video Generation},
  author = {He, Xuan and Jiang, Dongfu and Zhang, Ge and Ku, Max and Soni, Achint and Siu, Sherman and Chen, Haonan and Chandra, Abhranil and Jiang, Ziyan and Arulraj, Aaran and Wang, Kai and Do, Quy Duc and Ni, Yuansheng and Lyu, Bohan and Narsupalli, Yaswanth and Fan, Rongqi and Lyu, Zhiheng and Lin, Yuchen and Chen, Wenhu},
  journal = {ArXiv},
  year = {2024},
  volume={abs/2406.15252},
  url = {https://arxiv.org/abs/2406.15252},
}