Skip to content
项目

hysts/anime-face-detector

在 GitHub 查看

hysts/anime-face-detector has 548 GitHub stars and 44 forks. GitHot measured +1 stars in 24 hours, +3 in 7 days, and +5 in 15 days. Last checked 2026-11-22 13:06 UTC. Last push 2026-07-05.

这些数字如何统计

实测星标增长

24小时
+1
7天
+3
15天
+5
1个月
正在收集历史

各周期内的星标净变化,周期相互重叠,并非每日新增。缺失的测量保持空白。星标代表关注度,不代表软件质量。

关于

Anime Face Detector using mmdet and mmpose

机器学习

README.md

原文

Anime Face Detector

PyPI version Downloads Open In Colab Hugging Face Spaces MIT License GitHub stars

This is an anime face detector with facial landmark estimation, implemented in plain PyTorch. The face detectors (Faster R-CNN and YOLOv3) and the landmark model (HRNetV2, 28 points) were originally trained with mmdetection and mmpose; since v0.1.0 this package runs them without any OpenMMLab runtime dependency.

(To avoid copyright issues, the above demo uses images generated by the TADNE model.)

The model detects near-frontal anime faces and predicts 28 landmark points.

The result of k-means clustering of landmarks detected in real images:

The mean images of real images belonging to each cluster:

Installation

pip install anime-face-detector

Python 3.12 or later is required. PyTorch (2.2.0 or later) and torchvision are installed as dependencies; if you need a specific CUDA build, install them first by following the official instructions.

Usage

Open In Colab

import cv2

from anime_face_detector import create_detector

detector = create_detector("yolov3")
image = cv2.imread("assets/input.jpg")
preds = detector(image)
print(preds[0])
{'bbox': array([2.2450244e+03, 1.5940223e+03, 2.4116030e+03, 1.7458063e+03,
        9.9987185e-01], dtype=float32),
 'keypoints': array([[2.2593938e+03, 1.6680436e+03, 9.3236601e-01],
        [2.2825300e+03, 1.7051841e+03, 8.7208068e-01],
        [2.3412151e+03, 1.7281011e+03, 1.0052248e+00],
        [2.3941377e+03, 1.6825046e+03, 5.9705663e-01],
        [2.4039426e+03, 1.6541921e+03, 8.7139702e-01],
        [2.2625220e+03, 1.6330233e+03, 9.7608268e-01],
        [2.2804077e+03, 1.6408495e+03, 1.0021354e+00],
        [2.2969380e+03, 1.6494972e+03, 9.7812974e-01],
        [2.3357908e+03, 1.6453258e+03, 9.8418534e-01],
        [2.3475276e+03, 1.6355408e+03, 9.5060223e-01],
        [2.3612463e+03, 1.6262626e+03, 9.0553057e-01],
        [2.2682278e+03, 1.6631940e+03, 9.5465249e-01],
        [2.2814783e+03, 1.6616484e+03, 9.0782022e-01],
        [2.2987590e+03, 1.6692812e+03, 9.0256405e-01],
        [2.2833625e+03, 1.6879142e+03, 8.0303693e-01],
        [2.2934949e+03, 1.6909009e+03, 8.9718056e-01],
        [2.3021218e+03, 1.6863715e+03, 9.3882143e-01],
        [2.3471826e+03, 1.6636573e+03, 9.5727938e-01],
        [2.3677822e+03, 1.6540554e+03, 9.4890594e-01],
        [2.3889211e+03, 1.6611255e+03, 9.5125675e-01],
        [2.3575544e+03, 1.6800433e+03, 8.5919142e-01],
        [2.3688926e+03, 1.6800665e+03, 8.3275074e-01],
        [2.3804905e+03, 1.6761322e+03, 8.4160626e-01],
        [2.3165366e+03, 1.6947096e+03, 9.1840971e-01],
        [2.3282458e+03, 1.7104808e+03, 8.8045174e-01],
        [2.3380054e+03, 1.7114034e+03, 8.8357794e-01],
        [2.3485500e+03, 1.7080273e+03, 8.6284375e-01],
        [2.3378748e+03, 1.7118135e+03, 9.7880816e-01]], dtype=float32)}

Pretrained models

The pretrained weights are hosted on the Hugging Face Hub as safetensors files (faster-rcnn, yolov3, hrnetv2) and are downloaded automatically on first use. The original .pth checkpoints from the mmdetection/mmpose days remain available in the v0.0.1 release. The anime-face-detector-convert command converts such checkpoints, including self-trained ones, to safetensors.

Demo (using Gradio)

Hugging Face Spaces

Run locally

pip install anime-face-detector gradio
git clone https://github.com/hysts/anime-face-detector
cd anime-face-detector

python demo_gradio.py

License

The code in this repository is released under the MIT License (see LICENSE). The code under anime_face_detector/_vendor/ is derived from mmpose v0.29.0, mmdetection v2.28.2 and mmcv v1.7.0, which are licensed under the Apache License 2.0; see THIRD_PARTY_LICENSES.md for details.

Citation

If you find this repo useful for your research, please consider citing it:

@misc{anime-face-detector,
  author = {hysts},
  title = {Anime Face Detector},
  year = {2021},
  howpublished = {\url{https://github.com/hysts/anime-face-detector}}
}

Links

General

Anime face detection

Anime face landmark detection

Others

探索相关项目

加载中…