anime-face-detector

Why we included this project

Anime and illustrated content follows its own visual rules, and face detectors trained on photos often miss it. This package targets that gap: it detects near-frontal anime faces and returns 28 facial landmark points in one call. The detectors (Faster R-CNN and YOLOv3) and the landmark model (HRNetV2) were trained with mmdetection and mmpose, but since v0.1.0 the package runs them in plain PyTorch, so you get the pretrained weights without the OpenMMLab runtime dependency. A pip install and a few lines of Python are enough to get bounding boxes and landmarks, and the bundled Colab notebook makes it easy to check results quickly. It is a narrow, single-purpose tool rather than a general vision framework, which suits teams building anime-specific pipelines like face swapping or dataset preprocessing.