Python-人体姿势估计和跟踪的简单基线的TensorFlow实现

人体姿势估计和跟踪的简单基线的TensorFlow实现

#!/usr/bin/python3
# coding=utf-8

import os
import os.path as osp
import numpy as np
import cv2
import json
import pickle
import matplotlib.pyplot as plt

import sys
cur_dir = os.path.dirname（__file__）
sys.path.insert（0 osp.join（cur_dir ‘PythonAPI‘））
from pycocotools.coco import COCO
from pycocotools.cocoeval import COCOeval

class Dataset（object）:
    
    dataset_name = ‘COCO‘
    num_kps = 17
    kps_names = [‘nose‘ ‘l_eye‘ ‘r_eye‘ ‘l_ear‘ ‘r_ear‘ ‘l_shoulder‘
    ‘r_shoulder‘ ‘l_elbow‘ ‘r_elbow‘ ‘l_wrist‘ ‘r_wrist‘
    ‘l_hip‘ ‘r_hip‘ ‘l_knee‘ ‘r_knee‘ ‘l_ankle‘ ‘r_ankle‘]
    kps_symmetry = [（1 2） （3 4） （5 6） （7 8） （9 10） （11 12） （13 14） （15 16）]
    kps_lines = [（1 2） （0 1） （0 2） （2 4） （1 3） （6 8） （8 10） （5 7） （7 9） （12 14） （14 16） （11 13） （13 15） （5 6） （11 12）]

    human_det_path = osp.join（‘..‘ ‘data‘ dataset_name ‘dets‘ ‘human_detection.json‘） # human detection result
    img_path = osp.join（‘..‘ ‘data‘ dataset_name ‘images‘）
    train_annot_path = osp.join（‘..‘ ‘data‘ dataset_name ‘annotations‘ ‘person_keypoints_train2017.json‘）
    val_annot_path = osp.join（‘..‘ ‘data‘ dataset_name ‘annotations‘ ‘person_keypoints_val2017.json‘）
    test_annot_path = osp.join（‘..‘ ‘data‘ dataset_name ‘annotations‘ ‘image_info_test-dev2017.json‘）

    def load_train_data（self score=False）:
        coco = COCO（self.train_annot_path）
        train_data = []
        for aid in coco.anns.keys（）:
            ann = coco.anns[aid]
            imgname = ‘train2017/‘ + coco.imgs[ann[‘image_id‘]][‘file_name‘]
            joints = ann[‘keypoints‘]
 
            if （ann[‘image_id‘] not in coco.imgs） or ann[‘iscrowd‘] or （np.sum（joints[2::3]） == 0） or （ann[‘num_keypoints‘] == 0）:
                continue
           
            # sanitize bboxes
            x y w h = ann[‘bbox‘]
            img = coco.loadImgs（ann[‘image_id‘]）[0]
            width height = img[‘width‘] img[‘height‘]
            x1 = np.max（（0 x））
            y1 = np.max（（0 y））
            x2 = np.min（（width - 1 x1 + np.max（（0 w - 1））））
            y2 = np.min（（height - 1 y1 + np.max（（0 h - 1））））
            if ann[‘area‘] > 0 and x2 >= x1 and y2 >= y1:
                bbox = [x1 y1 x2-x1 y2-y1]
            else:
                continue
            
            if score:
                data = dict（image_id = ann[‘image_id‘] imgpath = imgname bbox=bbox joints=joints score=1）
            else:
                data = dict（image_id = ann[‘image_id‘] imgpath = imgname bbox=bbox joints=joints）

            train_data.append（data）

        return train_data
    
    def load_val_data_with_annot（self）:
        coco = COCO（self.val_annot_path）
        val_data = []
        for aid in coco.anns.keys（）:
            ann = coco.anns[aid]
            if ann[‘image_id‘] not in coco.imgs:
                continue
            imgname = ‘val2017/‘ + coco.imgs[ann[‘image_id‘]][‘file_name‘]
            bbox = ann[‘b

 属性            大小     日期    时间   名称
----------- ---------  ---------- -----  ----
     目录           0  2019-07-22 16:13  TF-SimpleHumanPose-master\
     文件          70  2019-07-22 16:13  TF-SimpleHumanPose-master\.gitignore
     文件       10132  2019-07-22 16:13  TF-SimpleHumanPose-master\README.md
     目录           0  2019-07-22 16:13  TF-SimpleHumanPose-master\data\
     目录           0  2019-07-22 16:13  TF-SimpleHumanPose-master\data\COCO\
     文件        6794  2019-07-22 16:13  TF-SimpleHumanPose-master\data\COCO\dataset.py
     目录           0  2019-07-22 16:13  TF-SimpleHumanPose-master\data\MPII\
     文件        4596  2019-07-22 16:13  TF-SimpleHumanPose-master\data\MPII\dataset.py
     目录           0  2019-07-22 16:13  TF-SimpleHumanPose-master\data\PoseTrack\
     文件        7295  2019-07-22 16:13  TF-SimpleHumanPose-master\data\PoseTrack\dataset.py
     目录           0  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\
     文件         110  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\Makefile
     文件          26  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\__init__.py
     目录           0  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\nets\
     文件           0  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\nets\__init__.py
     文件        8001  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\nets\basemodel.py
     文件       11902  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\nets\resnet_utils.py
     文件       13646  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\nets\resnet_v1.py
     目录           0  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\nms\
     文件           0  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\nms\__init__.py
     文件        2305  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\nms\cpu_nms.pyx
     文件         146  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\nms\gpu_nms.hpp
     文件        1179  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\nms\gpu_nms.pyx
     文件        3338  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\nms\nms.py
     文件        5062  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\nms\nms_kernel.cu
     文件        5184  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\nms\setup.py
     目录           0  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\tfflat\
     文件           0  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\tfflat\__init__.py
     文件       17243  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\tfflat\base.py
     文件       12741  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\tfflat\data_provider.py
     文件        1035  2019-07-22 16:13  TF-SimpleHumanPose-master\lib\tfflat\dpflow.py
............此处省略25个文件信息