VISTA: Enhancing Long-Duration and High-Resolution Video Understanding by Video Spatiotemporal Augmentation
This addresses data scarcity for video understanding in AI, though it is incremental as it builds on existing datasets and methods.
The paper tackled the challenge of processing long-duration and high-resolution videos in large multimodal models by proposing VISTA, a video spatiotemporal augmentation framework that synthesizes video instruction-following pairs, resulting in average improvements of 3.3% on long-video benchmarks and 6.5% on a new high-resolution benchmark.
Current large multimodal models (LMMs) face significant challenges in processing and comprehending long-duration or high-resolution videos, which is mainly due to the lack of high-quality datasets. To address this issue from a data-centric perspective, we propose VISTA, a simple yet effective Video Spatiotemporal Augmentation framework that synthesizes long-duration and high-resolution video instruction-following pairs from existing video-caption datasets. VISTA spatially and temporally combines videos to create new synthetic videos with extended durations and enhanced resolutions, and subsequently produces question-answer pairs pertaining to these newly synthesized videos. Based on this paradigm, we develop seven video augmentation methods and curate VISTA-400K, a video instruction-following dataset aimed at enhancing long-duration and high-resolution video understanding. Finetuning various video LMMs on our data resulted in an average improvement of 3.3% across four challenging benchmarks for long-video understanding. Furthermore, we introduce the first comprehensive high-resolution video understanding benchmark HRVideoBench, on which our finetuned models achieve a 6.5% performance gain. These results highlight the effectiveness of our framework.