行人检测
密集场景：
https://mp.weixin.qq.com/s?__biz=MzU4NTkwMDM1NA==&mid=2247485269&idx=1&sn=931314b297d93cebf4f3b732b9a42c42&chksm=fd82cf78caf5466ec6dc9aa457b619a1a8a28bb3bc5c9a5f30fd135bf5f052b04895e911ba12&mpshare=1&scene=1&srcid=&sharer_sharetime=1587697292880&sharer_shareid=f33650c82e6321f0f2bf3ea27a9b9ad3&key=f41bd6906825ad08795ef1ea07cb81f501edbecc91de1334d07772d9bd932795c0b14c1a0a1b9a621bb8233aac0beea801f637ed76ce7101a222526407d1fe065141e1fcb41806acf7ca3268a0333bd8&ascene=1&uin=MjIzODAyMTI0MA%3D%3D&devicetype=Windows+10&version=62080079&lang=zh_CN&exportkey=AQBOxtSoJzVI2loosGGf1Pw%3D&pass_ticket=ZTzQLkQUbHr9oTlO7J5bgBy1kgNI73%2F%2Bys3SQOUqgzjbTA7ZHdMduZ4tIp3exGjt
有人实现的：
https://github.com/Purkialo/CrowdDet
2018年 ALFnet
https://blog.csdn.net/jacke121/article/details/105721305
论文地址：
https://github.com/xingkongliang/Pedestrian-Detection
知乎专栏：
https://www.zhihu.com/topic/20087379/hot
torch版 轻量级RFB进行行人检测，AP达到0.7993，参数量仅有3.1MB，200 FPS
https://github.com/songwsx/RFSong-7993
tensorrt版：
https://github.com/xiaoxiaotao/person-detection
旷世研究院：CrowdHuman+Double Anchor
https://zhuanlan.zhihu.com/p/95253096
代码地址：
model:150M
https://github.com/liuwei16/CSP
We use the backbone ResNet-50 and MobileNet_v1
这个特别快，但是漏检很严重
https://github.com/YonghaoHe/A-Light-and-Fast-Face-Detector-for-Edge-Devices/tree/master/pedestrian_detection
2018的：tf的，有权重60多m，能测试(1024,512) 60ms
https://github.com/VideoObjectSearch/ALFNet
https://github.com/ysglh/ALFNet
16个月之前：
https://github.com/liuwei16/ALFNet
185M
https://github.com/open-mmlab/mmdetection 这个基础上的：
https://github.com/Leotju/MGAN
sklearn:效果不好
https://github.com/jacke121/Pedestrian-detection
torch版：backbone resnet50
https://github.com/lw396285v/CSP-pedestrian-detection-in-pytorch
数据集：
https://zhuanlan.zhihu.com/p/31836357
有数据集下载；
https://github.com/YonghaoHe/A-Light-and-Fast-Face-Detector-for-Edge-Devices/tree/master/pedestrian_detection