test-HunyuanVideo-pixelart-videos
收藏资源简介:
该数据集专注于动漫风格的像素艺术图像,这些图像经过精心挑选,以提升模型的性能。图像特点包括细节丰富、色彩鲜艳和构图良好。数据集是从大约100,000条Twitter帖子中筛选出的500个样本(图像+视频),通过自动过滤和手动筛选的过程收集。所有图像和视频都使用NVILA-15B模型进行了描述,并存储在`metadata.jsonl`文件中。数据集格式兼容ImageFolder和VideoFolder,便于使用。
This dataset focuses on anime-style pixel art images, which are meticulously selected to boost model performance. The images feature rich details, vibrant colors and well-composed compositions. The dataset includes 500 samples (images and videos) screened from approximately 100,000 Twitter posts, collected via a workflow combining automatic filtering and manual screening. All images and videos are annotated using the NVILA-15B model and stored in the `metadata.jsonl` file. The dataset format is compatible with ImageFolder and VideoFolder for ease of use.




