Instruct Video-to-Video (InsV2V)
收藏资源简介:
Instruct Video-to-Video (InsV2V) 数据集由亚马逊上海人工智能实验室创建,专注于视频到视频的转换任务。该数据集通过严格的文本和视频组件配对,为模型训练提供了理想的训练基础。数据集包含304,168对视频样本,每对包含一个输入视频及其编辑后的版本。InsV2V数据集的创建过程利用了大型语言模型和Prompt-to-Prompt方法,确保了视频内容与编辑指令之间的高度一致性。该数据集主要应用于文本驱动的视频编辑领域,旨在解决现有视频编辑方法中存在的资源密集型微调问题,提供一种更高效、用户友好的视频编辑解决方案。
The Instruct Video-to-Video (InsV2V) dataset was developed by Amazon Shanghai AI Laboratory, focusing on the video-to-video translation task. This dataset strictly pairs text and video components, serving as an ideal training foundation for model training. The dataset comprises 304,168 video sample pairs, each containing an input video and its edited counterpart. The development of the InsV2V dataset leverages large language models and the Prompt-to-Prompt method to ensure high consistency between video content and editing instructions. Primarily utilized in the domain of text-driven video editing, this dataset aims to resolve the resource-intensive fine-tuning challenge faced by existing video editing approaches, offering a more efficient and user-friendly video editing solution.




