3D_openpose/openpose_3d_2.py at master · RodrigoGantier/3D_openpose · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
import tensorflow as tf
import logging
import numpy as np
import cv2
import os
import glob
from slim_pose.network_cmu import CmuNetwork
from slim_pose.estimator import PoseEstimator
from slim_pose.lineal_model import lineal_model

from matplotlib import pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
import time

logger = logging.getLogger('train')
logger.setLevel(logging.DEBUG)
ch = logging.StreamHandler()
ch.setLevel(logging.DEBUG)
formatter = logging.Formatter('[%(asctime)s] [%(name)s] [%(levelname)s] %(message)s')
ch.setFormatter(formatter)
logger.addHandler(ch)


def post_est(outputs, batchsize, s=True):

    heat = outputs[:, :, :, :19]
    vect = outputs[:, :, :, 19:]
    human_parts = np.zeros([batchsize, 120, 18, 3])
    humnas_num = np.zeros([batchsize])
    for batch in range(batchsize):
        humans = PoseEstimator.estimate(heat[batch], vect[batch])
        humnas_num[batch] = len(humans)
        for h_num in range(len(humans)):
            for part_num in humans[h_num].body_parts.keys():
                x = humans[h_num].body_parts[part_num].x
                y = humans[h_num].body_parts[part_num].y
                score = humans[h_num].body_parts[part_num].score
                human_parts[batch, h_num, part_num, :] = np.array([y, x, score])
    if s:
        human_parts = human_parts[:, 0, :14, :2]
        human_parts = human_parts[:, :, ::-1].reshape(-1, 14 * 2)
    return human_parts.astype(np.float32), humnas_num.astype(np.int32)


def main():

    pretrain_path = '/media/rodrigo/c1d7e9c9-c8cb-402e-b241-9090925389b3/CMU_open_pose/tf-pose-estimation-master/models/numpy/openpose_coco.npy'
    batch_size = 1
    h_dim = 480
    w_dim = 480
    epochs = 10000

    img = tf.placeholder(tf.float32, [batch_size, h_dim, w_dim, 3])
    dropout = tf.placeholder_with_default(1.0, shape=())
    # human_3d_labels = tf.placeholder(tf.float32, [None, 14])
    # global_step = tf.Variable(0, trainable=False)

    # openpose net
    net = CmuNetwork({'image': img}, trainable=False)
    outputs = net.get_output()
    human_points, human_num = tf.py_func(post_est,
                                         [outputs, batch_size],
                                         [tf.float32, tf.int32],
                                         stateful=False, name='py_func')

    # linear net
    human_3d_predict = lineal_model(tf.reshape(human_points, [-1, 28]), dropout)

    config = tf.ConfigProto(allow_soft_placement=True, log_device_placement=False)
    with tf.Session(config=config) as sess:

        sess.run(tf.global_variables_initializer())
        tf.train.Saver().restore(sess,
                                 '/media/rodrigo/c1d7e9c9-c8cb-402e-b241-9090925389b3/human_action/test_tb/openpose_3d/save/save.ckpt')
        cap = cv2.VideoCapture(0)
        cap.set(3, w_dim)
        cap.set(4, h_dim)
        plt.ion()
        l1, l2 = [0, 1, 2, 3, 1, 5, 6, 1, 1, 8, 9, 11, 12], [1, 2, 3, 4, 5, 6, 7, 8, 11, 9, 10, 12, 13]
        fig = plt.figure()
        af = fig.add_subplot(121)
        ax = fig.add_subplot(122, projection='3d')
        # ax.view_init(-90, -90)

        for epoch in range(epochs):

                frames = np.zeros([batch_size, w_dim, h_dim, 3])

                for fr in range(batch_size):
                    ret_val, image = cap.read()
                    padd_d = (image.shape[0] - image.shape[1]) // 2
                    d_1, d_2 = abs(padd_d), 0
                    image = np.pad(image, [[d_1, d_1], [d_2, d_2], [0, 0]], mode='constant')
                    image = cv2.resize(image, (h_dim, w_dim))
                    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
                    frames[fr] = image

                kp_3d, kp_2d = sess.run([human_3d_predict, human_points], {img: frames})
                kp_2d = kp_2d.reshape([batch_size, 14, 2])
                kp_3d = np.append(kp_2d, kp_3d.reshape(batch_size, 14, 1), axis=2)
                for fr in range(batch_size):
                    af.cla()
                    af.imshow(frames[fr].astype(np.uint8))

                    ax.cla()
                    ax.set_xlabel('X Label')
                    ax.set_ylabel('Y Label')
                    ax.set_zlabel('Z Label')
                    # ax.view_init(-90, -90)
                    for ii in range(14):
                        xs, ys, zs = kp_3d[fr, ii]
                        if xs != 0:
                            ax.scatter(xs, ys, zs, c="r", marker="^")
                        for iii, pp in zip(l1, l2):
                            if kp_3d[fr, iii, 0] * kp_3d[fr, pp, 0] != 0:
                                ax.plot([kp_3d[fr, iii, 0], kp_3d[fr, pp, 0]],
                                        [kp_3d[fr, iii, 1], kp_3d[fr, pp, 1]],
                                        [kp_3d[fr, iii, 2], kp_3d[fr, pp, 2]],
                                        color='b')
                    plt.pause(0.01)
                    plt.draw()

    logger.info('Finish')

    print('finish')


if __name__ == '__main__':
    main()