VidCapBench: A Comprehensive Benchmark of Video Captioning for Controllable Text-to-Video Generation
This provides a tool for researchers and developers to improve controllable text-to-video models by better aligning captions with video content, though it is incremental as it builds on existing evaluation methods.
The paper tackles the lack of connection between video caption evaluation and text-to-video generation assessment by introducing VidCapBench, a benchmark that uses expert and human annotation to evaluate captions based on video attributes, showing superior stability and a positive correlation with T2V quality metrics.
The training of controllable text-to-video (T2V) models relies heavily on the alignment between videos and captions, yet little existing research connects video caption evaluation with T2V generation assessment. This paper introduces VidCapBench, a video caption evaluation scheme specifically designed for T2V generation, agnostic to any particular caption format. VidCapBench employs a data annotation pipeline, combining expert model labeling and human refinement, to associate each collected video with key information spanning video aesthetics, content, motion, and physical laws. VidCapBench then partitions these key information attributes into automatically assessable and manually assessable subsets, catering to both the rapid evaluation needs of agile development and the accuracy requirements of thorough validation. By evaluating numerous state-of-the-art captioning models, we demonstrate the superior stability and comprehensiveness of VidCapBench compared to existing video captioning evaluation approaches. Verification with off-the-shelf T2V models reveals a significant positive correlation between scores on VidCapBench and the T2V quality evaluation metrics, indicating that VidCapBench can provide valuable guidance for training T2V models. The project is available at https://github.com/VidCapBench/VidCapBench.