COCO 数据集 |
MPII数据集 | Posetrack数据集 | |
数据集格式 | COCO数据集的标注格式 | 0 - r ankle, 1 - r knee, 2 - r hip,3 - l hip,4 - l knee, 5 - l ankle, 6 - l ankle, 7 - l ankle,8 - upper neck, 9 - head top,10 - r wrist,11 - r elbow, 12 - r shoulder, 13 - l shoulder,14 - l elbow, 15 - l wrist |
0 - r ankle, 1 - r knee, 2 - r hip,3 - l hip,4 - l knee, 5 - l ankle, 8 - upper neck, 9 - head top,10 - r wrist,11 - r elbow, 12 - r shoulder, 13 - l shoulder,14 - l elbow, 15 - l wrist |
关键点数量 | 17 | 16 | 14 |
是否有mask | 有 | 无 | |
评估标准 |
|
||
state of art算法
|
目前主流的pose estimation方法:
pose-track和CPM的异同点:多了一个PAFs
目前存在的问题与常用解决思路:
现阶段有哪些大牛活跃在这个领域,并且目前的研究方向是什么:
COCO数据集keypoint标注:
{0-nose 1-Leye 2-Reye 3-Lear 4Rear 5-Lsho 6-Rsho 7-Lelb 8-Relb 9-Lwri 10-Rwri 11-Lhip 12-Rhip 13-Lkne 14-Rkne 15-Lank 16-Rank}
但是在 openpose 训练时,转换成了如下顺序:
{0-'nose', 1-'neck', 2-'Rsho', 3-'Relb', 4-'Rwri',5-'Lsho', 6-'Lelb', 7-'Lwri', 8-'Rhip', 9-'Rkne', 10-'Rank', 11-'Lhip', 12-'Lkne', 13-'Lank', 14-'Leye', 15-'Reye', 16-'Lear', 17-'Rear', 18-'pt19'}
其中 1-‘neck' 是openpose增加的一个,也能认为是人体中心点。。但是18-'pt19' 是背景类,最后会被剔除
%%MATLAB code
s_vec = net.forward(input_data); scores = cat(3, s_vec{2}(:,:,1:end-1), s_vec{1});
在Keypoint Annotations中,有个flag v 的定义: v=0 :没标记(x=y=0) || v=1 : 标记了但不可见 || v=2 : 标记了且可见
↑参考:http://blog.csdn.net/happyhorizion/article/details/77894205
MPII数据集标注:
{ 0 - r ankle, 1 - r knee, 2 - r hip,3 - l hip,4 - l knee, 5 - l ankle, 6 - pelvis, 7 - thorax,8 - upper neck, 9 - head top,10 - r wrist,11 - r elbow, 12 - r shoulder, 13 - l shoulder,14 - l elbow, 15 - l wrist }
openpose在训练时,剔除了6 - pelvis(盆骨), 7 - thorax(胸部)两个点,并增加了一个 center 点,共计输出是15个keypoint的heatmap和1个背景(要剔除)。
对应heatmap中的顺序为:{ 0-'nose', 1-'neck', 2-'Rsho', 3-'Relb', 4-'Rwri',5-'Lsho', 6-'Lelb', 7-'Lwri', 8-'Rhip', 9-'Rkne', 10-'Rank', 11-'Lhip', 12-'Lkne', 13-'Lank', 14-'center' }
14-'center' 这个点是由 2,3,12,13四个位置坐标取平均得到的!!!!!!!
else if(np == 43){ //int MPI_to_ours_1[15] = {9, 8,12,11,10,13,14,15, 2, 1, 0, 3, 4, 5, 7}; //int MPI_to_ours_2[15] = {9, 8,12,11,10,13,14,15, 2, 1, 0, 3, 4, 5, 6}; int MPI_to_ours_1[15] = {9, 8,12,11,10,13,14,15, 2, 1, 0, 3, 4, 5, 3}; int MPI_to_ours_2[15] = {9, 8,12,11,10,13,14,15, 2, 1, 0, 3, 4, 5, 2}; int MPI_to_ours_3[15] = {9, 8,12,11,10,13,14,15, 2, 1, 0, 3, 4, 5, 13}; int MPI_to_ours_4[15] = {9, 8,12,11,10,13,14,15, 2, 1, 0, 3, 4, 5, 12}; jo.joints.resize(np); jo.isVisible.resize(np); for(int i=0;i<15;i++){ // jo.joints[i] = (j.joints[MPI_to_ours_1[i]] + j.joints[MPI_to_ours_2[i]]) * 0.5; jo.joints[i] = (j.joints[MPI_to_ours_1[i]] + j.joints[MPI_to_ours_2[i]]+j.joints[MPI_to_ours_3[i]] + j.joints[MPI_to_ours_4[i]]) * 0.25; if(j.isVisible[MPI_to_ours_1[i]]==2 || j.isVisible[MPI_to_ours_2[i]]==2){ jo.isVisible[i] = 2; } else { jo.isVisible[i] = j.isVisible[MPI_to_ours_1[i]] && j.isVisible[MPI_to_ours_2[i]]; } } }
center map 是什么:
center map(绿色)是一个提前生成的高斯函数模板,用来把响应归拢到图像中心。长这样: