TachibanaYoshino/AnimeGAN
AnimeGAN
AnimeGAN 是一个使用 TensorFlow 开发的生成对抗网络（GAN）和图像转换器。它专为照片转动漫风格迁移而设计，利用深度学习系统将现实世界的照片和视频帧转换为动漫风格的图像。
该系统包括一个视频转动漫转换器，可在连续帧之间应用一致的视觉转换。它既支持在艺术数据集上训练生成网络以复制特定风格，也支持从检查点（Checkpoints）提取生成器权重以进行高效推理。
该项目提供图像优化工具，包括边缘平滑和模糊处理，以增强视觉过渡。它还通过连续帧流水线管理视频文件的处理。
Features
- Photo-to-Anime Converters - Converts real-world photographs into anime-style imagery using pre-trained generative networks.
- Anime Style Model Training - Trains generative networks on curated artistic datasets to replicate distinct anime visual styles.
- Generative Adversarial Networks - Employs a generative adversarial network architecture to synthesize images that mimic target anime styles.
- Image Stylization GANs - Implements a GAN architecture specifically designed to transform real-world photos into stylized anime images.
- Generative Adversarial Network Training - Trains generator and discriminator networks on artistic datasets to replicate specific anime visual aesthetics.
- Video-to-Video Stylization - Stylizes existing video footage into anime aesthetics using neural frame processing.
- Video-to-Anime Converters - Transforms motion pictures into anime animations while maintaining a consistent visual style across sequential frames.
- Generator-Only Inference - Isolates the image-generating network from the discriminator to perform fast anime style transformations.
- TensorFlow Model Inference - Executes predictions and image transformations using models trained within the TensorFlow ecosystem.
- Generator Weight Isolation - Isolates generator weights from trained checkpoints to enable lightweight style transfer inference.
- Sequential Frame Processing - Applies generative transformations across a timeline by processing video files as a sequence of individual frames.
