Reproduce “AIGCBench: Comprehensive Evaluation of Image-to-Video Content Generated by AI” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Fanda Fan, Chunjie Luo, Wanling Gao, Jianfeng Zhan

Researchers from the Chinese Academy of Sciences introduced AIGCBench, a comprehensive benchmark for Image-to-Video (I2V) content, integrating a diverse dataset and 11 evaluation metrics across four dimensions. It establishes a standardized framework for assessing I2V algorithms, with automated metrics demonstrating strong correlation with human judgment.

Key insights

  • Current I2V algorithms generally lack fine-grained control, struggling to generate video content with precise textual descriptions or specific motion directions.
  • Mainstream I2V models are limited in generating long videos (typically around 3 seconds), highlighting challenges in maintaining temporal consistency across extended sequences.
  • Video generation remains computationally intensive with slow inference speeds, requiring optimization for practical applications.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.