Technical Report: Competition Solution For Modelscope-Sora

TL;DR

This paper details the methodology used in the Modelscope-Sora challenge to improve video dataset quality through data processing techniques, aiming to enhance text-to-video generation performance under computational constraints.

Contribution

It introduces a comprehensive approach for data cleaning and processing tailored for video generation models in a competitive setting.

Findings

Improved dataset quality for video generation

Enhanced model performance in text-to-video tasks

Effective data filtering and acceleration techniques

Abstract

This report presents the approach adopted in the Modelscope-Sora challenge, which focuses on fine-tuning data for video generation models. The challenge evaluates participants' ability to analyze, clean, and generate high-quality datasets for video-based text-to-video tasks under specific computational constraints. The provided methodology involves data processing techniques such as video description generation, filtering, and acceleration. This report outlines the procedures and tools utilized to enhance the quality of training data, ensuring improved performance in text-to-video generation models.