-
Notifications
You must be signed in to change notification settings - Fork 0
/
dataset_posetrack.py
186 lines (140 loc) · 7.14 KB
/
dataset_posetrack.py
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
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
from __future__ import print_function, division
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.utils.data import Dataset, DataLoader
from torchvision.utils import save_image, make_grid
from torchvision import transforms, utils
import os
import cv2
import wandb
import random
import copy
import pandas as pd
import numpy as np
from tqdm import tqdm
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings("ignore")
# Code adjusted from: https://github.com/mkocabas/VIBE/blob/master/lib/data_utils/img_utils.py
def rotate_2d(pt_2d, rot_rad):
x = pt_2d[0]
y = pt_2d[1]
sn, cs = np.sin(rot_rad), np.cos(rot_rad)
xx = x * cs - y * sn
yy = x * sn + y * cs
return np.array([xx, yy], dtype=np.float32)
def gen_trans_from_patch_cv(c_x, c_y, src_width, src_height, dst_width, dst_height, scale, rot, inv=False):
# augment size with scale
src_w = src_width * scale
src_h = src_height * scale
src_center = np.zeros(2)
src_center[0] = c_x
src_center[1] = c_y # np.array([c_x, c_y], dtype=np.float32)
# augment rotation
rot_rad = np.pi * rot / 180
src_downdir = rotate_2d(np.array([0, src_h * 0.5], dtype=np.float32), rot_rad)
src_rightdir = rotate_2d(np.array([src_w * 0.5, 0], dtype=np.float32), rot_rad)
dst_w = dst_width
dst_h = dst_height
dst_center = np.array([dst_w * 0.5, dst_h * 0.5], dtype=np.float32)
dst_downdir = np.array([0, dst_h * 0.5], dtype=np.float32)
dst_rightdir = np.array([dst_w * 0.5, 0], dtype=np.float32)
src = np.zeros((3, 2), dtype=np.float32)
src[0, :] = src_center
src[1, :] = src_center + src_downdir
src[2, :] = src_center + src_rightdir
dst = np.zeros((3, 2), dtype=np.float32)
dst[0, :] = dst_center
dst[1, :] = dst_center + dst_downdir
dst[2, :] = dst_center + dst_rightdir
trans_inv = cv2.getAffineTransform(np.float32(dst), np.float32(src))
trans = cv2.getAffineTransform(np.float32(src), np.float32(dst))
return trans, trans_inv
def generate_image_patch(cvimg, c_x, c_y, bb_width, bb_height, patch_width, patch_height, do_flip, scale, rot):
img = cvimg.copy()
# c = center.copy()
img_height, img_width, img_channels = img.shape
if do_flip:
img = img[:, ::-1, :]
c_x = img_width - c_x - 1
trans, trans_inv = gen_trans_from_patch_cv(c_x, c_y, bb_width, bb_height, patch_width, patch_height, scale, rot, inv=False)
img_patch = cv2.warpAffine(img, trans, (int(patch_width), int(patch_height)), flags=cv2.INTER_LINEAR)
return img_patch, trans, trans_inv
class PoseTrack(Dataset):
"""Face Landmarks dataset."""
def __init__(self, window=1, frame_length=10, img_size=256, max_ids=5, given_id=-1, key_frame=True):
"""
Args:
csv_file (string): Path to the csv file with annotations.
root_dir (string): Directory with all the images.
transform (callable, optional): Optional transform to be applied
on a sample.
"""
self.mean = np.array([123.675, 116.280, 103.530])
self.std = np.array([58.395, 57.120, 57.375])
self.key_frame = key_frame
self.window = window
self.frame_length = frame_length
self.img_size = img_size
self.max_ids = max_ids
self.data = []
self.dataset = []
self.root_dir_2018 = "_DATA/Posetrack_2018/"
self.tracking_data = np.load("_DATA/posetrack_train.npz", allow_pickle=True)
for i_, track_ in enumerate(self.tracking_data['arr_0']):
for i in range((len(track_)-self.frame_length)//self.window + 1):
self.data.append(track_[i*self.window:(i*self.window)+frame_length])
def process_image(self, img, center, scale):
img, _, _ = generate_image_patch(img, center[0], center[1], scale, scale, self.img_size, self.img_size, False, 1.0, 0.0)
img = img[:, :, ::-1].copy().astype(np.float32)
img_n = img[:, :, ::-1].copy().astype(np.float32)
for n_c in range(3):
img_n[:, :, n_c] = (img_n[:, :, n_c] - self.mean[n_c]) / self.std[n_c]
return torch.from_numpy(np.transpose(img_n, (2, 0, 1)))
def __len__(self):
return len(self.data)
def __getitem__(self, idx):
img_names = []
ids = torch.zeros(self.frame_length, self.max_ids) + -1
centers = torch.zeros(self.frame_length, self.max_ids, 2)
scales = torch.zeros(self.frame_length, self.max_ids, 2)
bboxs = torch.zeros(self.frame_length, self.max_ids, 4)
keypoint_2d = torch.zeros(self.frame_length, self.max_ids, 15, 2)
keypoint_3d = torch.zeros(self.frame_length, self.max_ids, 15, 3)
keypoint_3t = torch.zeros(self.frame_length, self.max_ids, 15, 3)
pose_emb = torch.zeros(self.frame_length, self.max_ids, 2048)
apper_emb = torch.zeros(self.frame_length, self.max_ids, 512)
smpl_emb = torch.zeros(self.frame_length, self.max_ids, 226)
frames_ = self.data[idx]
for t, f in enumerate(frames_):
root_dir = self.root_dir_2018
img_name = os.path.join(root_dir , f[0])
img_names.append(img_name)
ids_ = f[1]
center_ = f[3]
scale_ = f[4]
bbox_ = f[5]
pose_ = f[6]
apper_ = f[7]
smpl_ = f[10]
for i, idx__ in enumerate(ids_[:self.max_ids]):
if(scale_[i][0]==0 or scale_[i][1]==0):
ids[t, i] = -1
continue
ids[t, i] = idx__
if(scale_[i][0]<50 or scale_[i][1]<100):
ids[t, i] = -1
image_width = f[9][i][1]
centers[t, i, :] = torch.from_numpy(center_[i] )
scales[t, i, :] = torch.from_numpy(scale_[i] )
bboxs[t, i, :] = torch.from_numpy(bbox_[i] )
keypoint_2d[t, i, :, :] = torch.from_numpy(f[8][0][i])
keypoint_3d[t, i, :, :] = torch.from_numpy(f[8][1][i])
keypoint_3t[t, i, :, :] = torch.from_numpy(f[8][2][i])
keypoint_3d[t, i, :, :] = keypoint_3t[t, i, :, :] + keypoint_3d[t, i, :, :]
pose_emb[t, i, :] = torch.from_numpy(pose_[i])
apper_emb[t, i, :] = torch.from_numpy(apper_[i])
smpl_emb[t, i, :] = torch.from_numpy(smpl_[i])
return 0, ids, centers, scales, bboxs, img_names, pose_emb, apper_emb, keypoint_3d, smpl_emb