// the find
youyuge34/Anime-InPainting
An application tool of edge-connect, which can do anime inpainting and drawing. 动漫人物图片自动修复，去马赛克，填补，去瑕疵
A thin application wrapper around the Edge-Connect GAN for anime image inpainting — lets you paint over damaged or censored regions of anime images and have the model fill them back in. It adds an OpenCV-based interactive GUI on top of the original research code, making it usable without writing any Python. Aimed at people who want to use the technique without touching the underlying model code.
The OpenCV GUI is a genuinely useful addition over raw Edge-Connect — mask drawing and edge editing in the same tool is practical. Auto-save/auto-load of checkpoint weights during training saves the usual manual babysitting. The config.yml approach for device selection (CPU vs GPU) is a small but sensible ergonomic improvement over hardcoded flags. Training manual is complete enough that someone could actually fine-tune on their own dataset.
Frozen since March 2019 — targets PyTorch 1.0, which is five major versions behind; expect breakage on any modern environment without significant dependency pinning work. The pretrained weights only cover anime faces, not full character images, which the README itself admits limits real-world usefulness. The author's own README points you to PI-REC as the better follow-up project, which is a strong signal this one is superseded. No tests, no CI, and the 'Todo' keys (0 and 1) in the UI were never implemented.