LSP数据集

下载地址：http://sam.johnson.io/research/
该数据集有两个类型，一个是聚焦单独人体，一个是包含人体所在环境的，根据需求下载
Leeds Sports Pose 是一个体育姿势数据集，其被分为竞技、羽毛球、棒球、体操、跑酷、足球、排球和网球几类，共包含约 2000 个姿势注释，图像均来自于 Flickr 的运动人员。

每张图像均为3通道的彩色图像，其像素的行范围为：[64,202],列范围为：[57,202]，并且每个图像都注有 14 个关节位置，其左侧和右侧关节始终以人为中心进行标记
1-14分别为： Right ankle、Right knee、Right hip、Left hip、Left knee、Left ankle、Right wrist、Right elbow、Right、shoulder、Left shoulder、Left elbow、Left wrist、Neck、Head top

数据集解析：

import glob
import os
from scipy.io import loadmat
import numpy as np
from PIL import Image
import cv2
 
colors = [(50, 0, 0), (100, 0, 0), (150, 0, 0), (200, 0, 0), (250, 0, 0), (0, 50, 0), (0, 100, 0),(0, 150, 0), (0, 200, 0),(0, 250, 0), (0, 0, 50), (0, 0, 100), (0, 0, 150), (0, 0, 200), (0, 0, 250), (50, 50, 50)]
 
def save_joints(mat_path,image_path,save_path,vis_save_path):
    """
    mat_path 是 lsp数据集mat文件所在地址，包含mat文件名
    image_path 是 lsp数据集图像的地址，不包含图像名
    save_path 是 要将lsp数据集中关键点保存的地址名
    vis_save_path是 可视化关键点保存的地址名
    lsp数据集共2000张图片
    """
    joints = loadmat(mat_path)
    joints = joints["joints"].transpose(2,0,1)
    joints = joints[:,:2,:]
 
    num = 0
    for img_path in glob.glob("%s/*.jpg" %image_path):
        img_name = img_path.split("\\")[-1]
        vissavepath = vis_save_path + img_name
        img = Image.open(img_path)
        img = np.array(img,dtype=np.uint8)
        img = cv2.cvtColor(img,cv2.COLOR_RGB2BGR)
        cen_points = joints[num,...]
        points_num = cen_points.shape[-1]
        point_dict = {}
        for points_ in range(points_num):
            point_x = cen_points[0,points_]
            point_y = cen_points[1,points_]
            point_dict[str(points_)] = [int(point_x),int(point_y)]
            img1 = cv2.circle(img, (int(point_x), int(point_y)), 5, colors[points_], 
                              thickness=-1)
            img1 = cv2.putText(img, str(points_),
                               (int(point_x) + 10, int(point_y)),
                                cv2.FONT_HERSHEY_SIMPLEX, 1, colors[points_], 1)
 
        with open(os.path.join(save_path,img_name.split(".")[0]+".txt"),"w") as img_txt:
            img_txt.write(str(point_dict))
        img_txt.close()
        num += 1
        # 若不想看图片中关键点的位置是否准确，请注释掉后面两行
        # cv2.imshow("img",img)
        # cv2.waitKey()
        cv2.imwrite(vissavepath, img)


mat_path = r'E:\lap\lsp_dataset_original\joints.mat'
image_path = r'E:\lap\lsp_dataset_original\images'
save_path = r'E:\lap\lsp_dataset_original\txt'
vis_save_path =r'E:\lap\lsp_dataset_original\vis\\'
save_joints(mat_path,image_path,save_path,vis_save_path)

全场景：

单独人体：

参考：https://blog.csdn.net/weixin_44743827/article/details/123523924