T2V-CompBench
收藏资源简介:
T2V-CompBench是由香港大学、香港中文大学和华为诺亚方舟实验室联合创建的一个综合性文本到视频生成基准数据集。该数据集包含700个文本提示,涵盖了七个类别,如一致属性绑定、动态属性绑定、空间关系等,旨在评估模型在复杂场景中生成视频的能力。数据集的创建过程中,特别强调了时间动态和动作动词的使用,确保每个提示至少包含一个动作动词。T2V-CompBench主要应用于文本到视频生成模型的评估和研究,旨在解决模型在组合生成方面的挑战。
T2V-CompBench is a comprehensive text-to-video generation benchmark dataset jointly created by The University of Hong Kong, The Chinese University of Hong Kong, and Huawei Noah's Ark Lab. This dataset contains 700 text prompts covering seven categories including consistent attribute binding, dynamic attribute binding, spatial relations and others, aiming to evaluate the video generation capabilities of models in complex scenarios. During the creation of the dataset, special emphasis was placed on temporal dynamics and the use of action verbs, ensuring that each prompt includes at least one action verb. T2V-CompBench is mainly applied to the evaluation and research of text-to-video generation models, with the purpose of addressing the challenges in compositional generation faced by such models.

- 1T2V-CompBench: A Comprehensive Benchmark for Compositional Text-to-video Generation香港大学, 香港中文大学, 华为诺亚方舟实验室 · 2024年



