人头检测学习笔记

 

yolov5

c++ tensorrt 有模型：

https://github.com/RichardoMrMu/yolov5-head-detect

python版：人体检测，不是人头检测。

https://github.com/RichardoMrMu/yolov5-head-detector

brainwash数据集

网盘链接：https://pan.baidu.com/s/1Vgr6jZByU41TPd2tkiPMwA 提取码：rnk3 

有数据集，不能访问了

https://drive.google.com/open?id=1zn-AGmsBqVheFPnDTXWBpeo3XRH1Ho15

https://drive.google.com/open?id=1LiTDMWk0KglGueJCaxgneEA_ltvEbUDV

后来发现就是这个数据集：

YOLOv4人头检测器训练数据集_人头检测数据集,yolov5人头检测-Linux文档类资源-CSDN下载

https://github.com/kly1997/head_shoulder-detection-by-yolov3

YOLOv4人头检测器训练数据集_人头检测数据集,yolov5人头检测-Linux文档类资源-CSDN下载

人头检测：

GitHub - tgisaturday/S3FD_ATSS_SAPD: Applying Adaptive Training Sample Selection and Soft Anchor Point Detection to S3FD

 gpu下90ms左右，在家测试有误检，自己的电脑上。

sfd_head_30000.pth

自己整理的摄像头测试代码： 

# -*- coding:utf-8 -*-

from __future__ import division
from __future__ import absolute_import
from __future__ import print_function

import os
import torch
import argparse
import torch.nn as nn
import torch.utils.data as data
import torch.backends.cudnn as cudnn
import torchvision.transforms as transforms

import cv2
import time
import numpy as np
from PIL import Image

from datasets.config import cfg
from s3fd import build_s3fd
from torch.autograd import Variable
from utils.augmentations import to_chw_bgr

from alfred.dl.torch.common import device

parser = argparse.ArgumentParser(description='s3df demo')
parser.add_argument('--save_dir', type=str, default='tmp/', help='Directory for detect result')
parser.add_argument('--model', type=str, default='./sfd_head_30000.pth', help='trained model')
parser.add_argument('--thresh', default=0.4, type=float, help='Final confidence threshold')
args = parser.parse_args()

use_cuda = torch.cuda.is_available()

torch.set_default_tensor_type('torch.FloatTensor')
if use_cuda:
    torch.set_default_tensor_type('torch.cuda.FloatTensor')

def detect(net, thresh):

    cap=cv2.VideoCapture(0)

    while True:

        ret,img=cap.read()
        if img is not None:
            img=cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
            height, width, _ = img.shape
            # image = cv2.resize(img, (960, 680))
            x = to_chw_bgr(img)
            x = x.astype('float32')
            x -= cfg.img_mean
            x = x[[2, 1, 0], :, :]

            with torch.no_grad():
                x = torch.from_numpy(x).unsqueeze(0).to(device)
                t1 = time.time()
                y = net(x)
                detections = y.data
                scale = torch.Tensor([img.shape[1], img.shape[0], img.shape[1], img.shape[0]])

                for i in range(detections.size(1)):
                    j = 0
                    while detections[0, i, j, 0] >= thresh:
                        score = detections[0, i, j, 0]
                        pt = (detections[0, i, j, 1:] * scale).cpu().numpy()
                        left_up, right_bottom = (int(pt[0]), int(pt[1])), (int(pt[2]), int(pt[3]))
                        j += 1
                        cv2.rectangle(img, left_up, right_bottom, (0, 0, 255), 2)
                        conf = "{:.3f}".format(score)
                        point = (int(left_up[0]), int(left_up[1] - 5))
                        cv2.putText(img, conf, point, cv2.FONT_HERSHEY_COMPLEX,            0.6, (0, 255, 0), 1)

                t2 = time.time()
                print('  timer:{}',t2 - t1)

                cv2.imshow('rr', img)
                cv2.waitKey(1)


if __name__ == '__main__':
    net = build_s3fd('test', cfg.NUM_CLASSES)
    net.load_state_dict(torch.load(args.model))
    net.eval()

    if use_cuda:
        net.cuda()
        cudnn.benckmark = True

    detect(net, args.thresh)

数据集：

 SCUT-HEAD

下载地址：

GitHub - Master-cai/SCUT-HEAD-Dataset-Release: SCUT HEAD is a large-scale head detection dataset, including 4405 images labeld with 111251 heads.

An implimentation of yoloV3 in head detection ,keras and tensorflow backend used

This repository is based on qqwweee/keras-yolo3.We train the keras yolov3 model in brainwash dataset,and make detection on human heads by this model.

人头数据集2：

人头检测训练数据集（带标签）_人头检测数据集,人头数据集-互联网文档类资源-CSDN下载

快速精准的人头检测，有数据，有代码

https://github.com/aditya-vora/FCHD-Fully-Convolutional-Head-Detector

修改的代码：加载预训练改了

head_detector = Head_Detector_VGG16(ratios=[1], anchor_scales=[2,4])
dict = torch.load(opt.caffe_pretrain_path)
head_detector.load_state_dict(dict['model'])

权重66m，有漏检，值得看看

mtcnn: 人脸检测，不是人头检测

https://github.com/Seanlinx/mtcnn

无源码：

https://github.com/samylee/HeadDetect_MTCNN

有人头数据集

https://github.com/zuoqing1988/train-mtcnn-head

有漏捡：

https://github.com/wnov/Keras_Head_Detection_YoloV3

人头和眼睛：

人头眼睛检测代码-C/C++代码类资源-CSDN下载

这个效果还行 keras，能检测安全帽，有误检

https://github.com/jacke121/head-detection-using-yolo

读取时加一行代码：

weights=np.insert(weights, 0, values=weights[0], axis=0)

人体检测，不是人头检测

GitHub - xinsuinizhuan/yolov5-head-detector: A python implementation of yolov5 and deepsort

 B超图：

GitHub - Rippumokudo29/detect_head_with_yolov5: The system to detect fetal heads in input images, using yolov5