H36M-Toolbox/norm.py at main · Visual-Pose-Lab/H36M-Toolbox · 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
import numpy as np
import os.path as osp
from scipy.io import loadmat
from subprocess import call
from os import makedirs
import pickle
from img import crop_image
import cv2
import os
from tqdm import tqdm

camera_index = {
    0:54138969,
    1:55011271,
    2:58860488,
    3:60457274
}

image_shape = (192, 256)

from metadata import load_h36m_metadata
metadata = load_h36m_metadata()


def _infer_box(pose3d, camera, rootIdx):
    root_joint = pose3d[rootIdx, :]
    tl_joint = root_joint.copy()
    tl_joint[0] -= 1000.0
    tl_joint[1] -= 900.0
    br_joint = root_joint.copy()
    br_joint[0] += 1000.0
    br_joint[1] += 1100.0
    tl_joint = np.reshape(tl_joint, (1, 3))
    br_joint = np.reshape(br_joint, (1, 3))

    tl2d = _weak_project(tl_joint, camera['fx'], camera['fy'], camera['cx'],
                         camera['cy']).flatten()

    br2d = _weak_project(br_joint, camera['fx'], camera['fy'], camera['cx'],
                         camera['cy']).flatten()
    return np.array([tl2d[0], tl2d[1], br2d[0], br2d[1]])


def _weak_project(pose3d, fx, fy, cx, cy):
    pose2d = pose3d[:, :2] / pose3d[:, 2:3]
    pose2d[:, 0] *= fx
    pose2d[:, 1] *= fy
    pose2d[:, 0] += cx
    pose2d[:, 1] += cy
    return pose2d

makedirs('/home/ltf/dataset/croped', exist_ok=True)

def normalize_screen_coordinates(X, w, h):
    assert X.shape[-1] == 2

    # Normalize so that [0, w] is mapped to [-1, 1], while preserving the aspect ratio
    return X/w*2 - [1, h/w]

# cnt = 0

# print('Loading 2D detections...')
pose2dgt = np.load('/home/ltf/data/xw/MHFormer/dataset/h36m/data_2d_h36m_gt.npz', allow_pickle=True)
pose2dgt_metadata = pose2dgt['metadata'].item()
pose2dgt = pose2dgt['positions_2d'].item()


pose2dcpn = np.load('/home/ltf/data/xw/MHFormer/dataset/h36m/data_2d_h36m_cpn_ft_h36m_dbb.npz', allow_pickle=True)
pose2dcpn = pose2dcpn['positions_2d'].item()

image_paths = np.load('crop_image_paths.npz', allow_pickle=True)
image_paths = image_paths['image_path'].item()

pose2dgt_norm = {}
pose2dcpn_norm = {}

for sub in pose2dgt.keys():
    pose2dgt_norm[sub] = {}
    pose2dcpn_norm[sub] = {}
    for act in pose2dgt[sub].keys():
        pose2dgt_norm_ = []
        pose2dcpn_norm_ = []
        for ca in range(len(pose2dgt[sub][act])):
            nposes = len(pose2dgt[sub][act][ca])
            print('crop image on subject: {} action: {} view: {}'.format(sub, act, ca))

            image_path = image_paths[sub][act][ca][0]
            image = cv2.imread(os.path.join('/home/ltf/dataset/', image_path), cv2.IMREAD_COLOR | cv2.IMREAD_IGNORE_ORIENTATION)
            h, w, _ = image.shape

            pose2dgt_norm_.append(normalize_screen_coordinates(pose2dgt[sub][act][ca][..., :2], w=w, h=h))
            pose2dcpn_norm_.append(normalize_screen_coordinates(pose2dcpn[sub][act][ca][..., :2], w=w, h=h))
        pose2dgt_norm[sub][act] = pose2dgt_norm_
        pose2dcpn_norm[sub][act] = pose2dcpn_norm_


print('Saving...')
metadata = {
    'num_joints': pose2dgt_metadata['num_joints'],
    'keypoints_symmetry': pose2dgt_metadata['keypoints_symmetry']
}
np.savez_compressed('/home/ltf/dataset/data_2d_h36m_gt_norm.npz', positions_2d=pose2dgt_norm, metadata=metadata)
np.savez_compressed('/home/ltf/dataset/data_2d_h36m_cpn_norm.npz', positions_2d=pose2dcpn_norm, metadata=metadata)

print('Done.')