Style30K Illusion Dataset
收藏资源简介:
Style30K Illusion Dataset是一个用于视频风格迁移和风格化生成的数据集,由快手科技和香港科技大学联合创建。该数据集包含30000张风格图像,分为约30个风格组,旨在通过对比学习提升风格提取的准确性。数据集通过模型幻觉技术生成,确保了风格一致性,避免了手动收集和分组的繁琐过程。该数据集主要应用于视频风格迁移和风格化生成任务,旨在解决现有方法在风格一致性和内容泄露方面的不足。
Style30K Illusion Dataset is a dataset for video style transfer and stylized generation, jointly created by Kuaishou Technology and The Hong Kong University of Science and Technology. It contains 30,000 stylized images divided into approximately 30 style groups, aiming to improve the accuracy of style extraction through contrastive learning. The dataset is generated via model hallucination technology, ensuring style consistency and eliminating the tedious process of manual collection and grouping. It is mainly applied to video style transfer and stylized generation tasks, aiming to address the shortcomings of existing methods in terms of style consistency and content leakage.

- 1StyleMaster: Stylize Your Video with Artistic Generation and Translation香港科技大学, 快手科技 · 2024年



