xavierjiezou / anime-face-detection

Summary of anime face detection methods based on Python.

Introduction

This repository is the summary of anime face detection methods based on Python.

Install

There are four methods for anime face detection: lbp, mlp, hog and ssd.

• lbp: pip install opencv-python
• mlp: pip install pillow opencv-python animeface
• hog: pip install opencv-python dlib
• ssd: pip install opencv-python numpy torch

Usage

For image detection:

python ./image_detect.py ./test/1.jpg lbp

For video detection:

python ./video_detect.py ./test/1.mp4 lbp

• Param one: path of images/videos
• Param two: method to be selected

Test Sample

Test Device

• CPU：12 Intel(R) Xeon(R) CPU E5-2603 v4 @ 1.70GHz
• GPU：8 NVIDIA GeForce GTX 1080 Ti

Anime Face Detection

Anime Face Detection Based on LBP

Repository

https://github.com/nagadomi/lbpcascade_animeface

Environment

• Module: pip install opencv-python
• Model: lbp_anime_face_detect.xml

Example

lbp_anime_face_detect.py

import cv2
import sys


def lbp_anime_face_detect(file_name):
    img = cv2.imread(file_name)
    img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    img_gray = cv2.equalizeHist(img_gray) 
    face_cascade = cv2.CascadeClassifier('../model/lbp_anime_face_detect.xml')
    faces = face_cascade.detectMultiScale(img_gray)
    for x, y, w, h in faces:
        img = cv2.rectangle(img, (x, y), (x+w, y+h), (255, 0, 255), 5)
    cv2.imwrite(f'../result/lbp_anime_face_detect_{file_name[-5]}.jpg', img) 


if __name__ == '__main__':
    lbp_anime_face_detect(sys.argv[1])

Result

Total	Missing	Error	Time
13	1	1	1.20s

Anime Face Detection Based on MLP

Repository

https://github.com/nya3jp/python-animeface

Environment

• System Requirements: Only for Linux system
• Module: pip install pillow opencv-python animeface

---

Note: If pip install animeface reports an error, please download the compiled file animeface-1.1.0-cp37-cp37m-manylinux1_x86_64.whl.whl, and then use the following command to install: (It is mandatory that the Python version is 3.7)

pip install animeface-1.1.0-cp37-cp37m-manylinux1_x86_64.whl

Example

mlp_anime_face_detect.py

import cv2
import animeface
from PIL import Image
import sys


def mlp_anime_face_detect(file_name):
    img = cv2.imread(file_name)  
    img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)  
    img_gray = cv2.equalizeHist(img_gray)  
    faces = animeface.detect(Image.fromarray(img_gray)) 
    for each in faces:  
        temp = each.face.pos
        x = temp.x
        y = temp.y
        w = temp.width
        h = temp.height
        img = cv2.rectangle(img, (x, y), (x+w, y+h), (255, 0, 255), 5)  
    cv2.imwrite(f'../result/mlp_anime_face_detect_{file_name[-5]}.jpg', img)  


if __name__ == '__main__':
    mlp_anime_face_detect(sys.argv[1])

Result

Total	Missing	Error	Time
17	0	4	28.28s

Anime Face Detection Based on HOG

Repository

https://github.com/marron-akanishi/AFD

Environment

• Module: pip install opencv-python dlib
• Model: hog_anime_face_detect.svm

---

Note：The dlib library needs to be compiled with the C++ compiler after downloading, so you need to install Visual Studio and configure the C++ compilation environment. If C++ environment have been installed and configured, please ignore; If not, please see this Article.

Example

hog_anime_face_detect.py

import cv2
import dlib
import sys


def hog_anime_face_detect(file_name):
    img = cv2.imread(file_name) 
    img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) 
    img_gray = cv2.equalizeHist(img_gray)  
    face_detector = dlib.simple_object_detector('../model/hog_anime_face_detect.svm') 
    faces = face_detector(img_gray)
    for face in faces: 
        left = face.left()
        top = face.top()
        right = face.right()
        bottom = face.bottom()
        cv2.rectangle(img, (left, top), (right, bottom), (255, 0, 255), 5) 
    cv2.imwrite(f'../result/hog_anime_face_detect_{file_name[-5]}.jpg', img)  


if __name__ == '__main__':
    hog_anime_face_detect(sys.argv[1])

Result

Total	Missing	Error	Time
10	3	0	2.42s

Anime Face Detection Based on SSD

Repository

https://github.com/WynMew/AnimeFaceBoxes

Environment

• Module: pip install opencv-python numpy torch
• Model: ssd_anime_face_detect.pth

Example

ssd_anime_face_detect.py (The code is too long to display, Please download and view)

Result

Total	Missing	Error	Time
13	0	0	0.72s

Other Experiment

Test Sample

【Painter】ゆりりん【Pixiv ID】88049646

【Painter】Nahaki【Pixiv ID】84678881

【Painter】A.one【Pixiv ID】66996496

【Painter】D.【Pixiv ID】68074033

Test Result

Analysis

The anime face detection algorithm based on MLP is too slow to meet the requirements of practical applications. The other three algorithms behave differently on different sample image.

Future Work

• Anime Face Detection Based on Faster-RCNN: https://github.com/qhgz2013/anime-face-detector/
• Anime Face Detection Based on CNN: https://github.com/ShiqiYu/libfacedetection
• Other Related Work: https://github.com/search?p=1&q=anime+face+detection&type=Repositories

Cite

https://github.com/nagadomi/lbpcascade_animeface

https://github.com/nya3jp/python-animeface

https://github.com/marron-akanishi/AFD

https://github.com/WynMew/AnimeFaceBoxes

https://github.com/hiromu/AnimeFace