[CV] Detect face with mask
疫情期间,有公司已经做出了"谁没带口罩上街"的检测模型。有点意思,出于兴趣,了解一下相关的技术。
学习
数据集
Ref: FDDB: Face Detection Data Set and Benchmark
Ref: WIDER FACE: A Face Detection Benchmark
Ref: 口罩遮挡人脸数据集:Real-World Masked Face Dataset,简称 RMFD
相关资源
人脸特征点:[Face] Outline - Key features detection of human face
CMU Github: openpose
[Face] Outline - Key features detection of human face
[CV Timeline] Applications of Face recognition and PoseNet
戴口罩
From: 疫情加速百度人脸识别变革:戴口罩也能准确识别,迅速上线,iPhone可以抄作业了
1. 基于空间位置的注意力机制特征学习
2. 特征可视化结果
3. 合成口罩图片
模型方案
From: Masked Face Recognition Dataset and Application
From: Detecting Masked Faces in the Wild With LLE-CNNs
LLE-CNN 主页:http://www.escience.cn/people/geshiming/mafa.html [页底的 Github 包含了大量相关模型链接]
根据2018年数据:in Intelligent Computing Methodologies: 15th International Conference, ICIC 2019
MobileNet分类模型可以

工程上,检测模型要慢上3-6倍。

实践
OpenPose on OpenCV
Ref: OpenPose 基于OpenCV DNN 的多人姿态估计 --> https://github.com/spmallick/learnopencv/tree/master/OpenPose-Multi-Person
Ref: OpenPose 基于OpenCV DNN 的多人姿态估计
Code: https://github.com/spmallick/learnopencv/tree/master/OpenPose-Multi-Person [valid]
Model download: https://github.com/CMU-Perceptual-Computing-Lab/openpose/tree/master/models/pose/coco
编译指令:
g++ -o3 -std=c++11 multi-person-openpose.cpp `pkg-config --libs --cflags opencv` -lpthread -o multi-person-openpose
就是有点慢,2ms, opencv3.4.5
OpenPose on TF
使用TF会如何呢:[CV] OpenPose on TensorFlow
/* continue */

浙公网安备 33010602011771号