face2anime

该数据集包含了多样化的动漫风格，既包括动漫人脸也包括自然照片人脸。具体来说，从Danbooru2019数据集中收集了8,898张动漫人脸，并从CelebA-HQ数据集中随机选取了8,898张女性人脸。在规模上，两个领域各有8,000张训练图像和898张测试图像，总计17,796张图像。该数据集的任务是进行从自然人脸到动漫人脸的图像转换。

This dataset features diverse anime art styles, including both anime faces and natural photographic human faces. Specifically, 8,898 anime faces were collected from the Danbooru2019 dataset, and 8,898 female human faces were randomly selected from the CelebA-HQ dataset. In terms of scale, each domain has 8,000 training images and 898 test images, totaling 17,796 images. The task of this dataset is image translation from natural human faces to anime faces.

搜集汇总

背景与挑战

背景概述

face2anime是一个专门用于无监督动漫人脸生成的数据集，包含17,796张图像，分为8,898张动漫人脸和8,898张自然照片人脸，所有图像均对齐并裁剪为128×128大小。该数据集比类似数据集更大且更具多样性，涵盖了多种动漫风格（如面部姿势、绘画风格、颜色、发型等），旨在支持生成对抗网络（GAN）模型的训练和评估。