-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathtest_image.py
268 lines (214 loc) · 10.1 KB
/
test_image.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
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
from collections import OrderedDict
from nets import Model
from utils import inference
import argparse
import torch
from file import MkdirSimple, match_images
import os
from tqdm.contrib import tzip
from PIL import Image
import numpy as np
from time import time
import cv2
from file import Walk, MkdirSimple
DATA_TYPE = ['kitti', 'dl', 'depth', 'server']
DIR_SHAPE = [
['left', 'right'],
['image_02', 'image_03'],
['cam0', 'cam1'],
['L', 'R']
]
def GetArgs():
parser = argparse.ArgumentParser(description='LaC')
parser.add_argument('--no_cuda', action='store_true', default=False)
parser.add_argument('--ignore_target', action='store_true', default=False)
parser.add_argument('--gpu_id', type=str, default='0')
parser.add_argument('--data_path', type=str, default='/media/data/dataset/KITTI/data_scene_flow/training/')
parser.add_argument('--model_file', type=str, default='models/crestereo_eth3d.pth')
parser.add_argument('--output', type=str)
parser.add_argument('--bf', type=float, default=14.2, help="baseline length multiply focal length")
parser.add_argument('--max_depth', type=int, default=1000, help="the valide max depth (cm)")
parser.add_argument('--scale', type=float, default=1, help="scale image to super resolution")
args = parser.parse_args()
return args
def GetImages(path):
matched_lists = None
for l, r in DIR_SHAPE:
left = os.path.join(path, l)
right = os.path.join(path, r)
if os.path.exists(left) and os.path.exists(right):
matched_lists = match_images([left, right])
break
return matched_lists[0], matched_lists[1], len(left.rstrip('/'))
def disp_to_depth(disp, min_depth, max_depth):
"""Convert network's sigmoid output into depth prediction
The formula for this conversion is given in the 'additional considerations'
section of the paper.
"""
min_disp = 1 / max_depth
max_disp = 1 / min_depth
scaled_disp = min_disp + (max_disp - min_disp) * disp
depth = 1 / scaled_disp
return scaled_disp, depth
def GetDepthImg(img):
depth_img_rest = img.copy()
depth_img_R = depth_img_rest.copy()
depth_img_R[depth_img_rest > 255] = 255
depth_img_rest[depth_img_rest < 255] = 255
depth_img_rest -= 255
depth_img_G = depth_img_rest.copy()
depth_img_G[depth_img_rest > 255] = 255
depth_img_rest[depth_img_rest < 255] = 255
depth_img_rest -= 255
depth_img_B = depth_img_rest.copy()
depth_img_B[depth_img_rest > 255] = 255
depth_img_rgb = np.stack([depth_img_R, depth_img_G, depth_img_B], axis=2)
return depth_img_rgb.astype(np.uint8)
def WriteDepth(predict_np, limg, path, name, bf, max_value=65535):
name = os.path.splitext(name)[0] + ".png"
name = name.replace('left', 'depth').replace('right', 'depth').replace('cam0', 'depth').replace('cam1', 'depth')
output_concat_color = os.path.join(path, "rgb_color", name)
output_concat_gray = os.path.join(path, "rgb_depth", name)
output_concat_depth = os.path.join(path, "rgb_blue", name)
output_concat = os.path.join(path, "concat", name)
output_gray = os.path.join(path, "depth", name)
output_float = os.path.join(path, "tiff", os.path.splitext(name)[0] + ".tiff")
output_depth = os.path.join(path, "depth_blue", name)
output_color = os.path.join(path, "color", name)
MkdirSimple(output_gray)
# MkdirSimple(output_float)
if bf > 0:
depth_img = bf / predict_np * 100 # to cm
depth_img = np.clip(depth_img, 0, max_value)
depth_img_u16 = depth_img / max_value * 65535
depth_img_u16 = depth_img_u16.astype("uint16")
else:
depth_img_u16 = predict_np
depth_img = depth_img_u16 / 65535 * 2500
cv2.imwrite(output_gray, depth_img_u16)
# return
invalid = depth_img < 0.001
depth_norm = np.clip(depth_img, 1, max_value)
depth_norm = 1.0 / depth_norm
depth_norm[invalid] = 0
depth_norm = (depth_norm - depth_norm.min()) / (depth_norm.max() - depth_norm.min()) * 255.0
depth_norm = depth_norm.astype(np.uint8)
if bf < 0:
predict_np = depth_norm
color_img = cv2.applyColorMap(depth_norm, cv2.COLORMAP_MAGMA)
limg_cv = limg # cv2.cvtColor(np.asarray(limg), cv2.COLOR_RGB2BGR)
concat_img_color = np.vstack([limg_cv, color_img])
# todo: hao 2025-02-10 14:13 - if input depth image
predict_np_rgb = np.stack([predict_np, predict_np, predict_np], axis=2)
concat_img_gray = np.vstack([limg_cv, predict_np_rgb])
# get depth
depth_img_rgb = GetDepthImg(depth_img)
concat_img_depth = np.vstack([limg_cv, depth_img_rgb])
concat = np.hstack([np.vstack([limg_cv, color_img]), np.vstack([predict_np_rgb, depth_img_rgb])])
# MkdirSimple(output_concat_color)
# MkdirSimple(output_concat_gray)
# MkdirSimple(output_concat_depth)
# MkdirSimple(output_depth)
# MkdirSimple(output_color)
MkdirSimple(output_concat)
# cv2.imwrite(output_concat_color, concat_img_color)
# cv2.imwrite(output_concat_gray, concat_img_gray)
# cv2.imwrite(output_color, color_img)
# cv2.imwrite(output_depth, depth_img_rgb)
# cv2.imwrite(output_concat_depth, concat_img_depth)
cv2.imwrite(output_concat, concat)
def main():
args = GetArgs()
bf = args.bf * args.scale
# output_dir = os.path.join(args.output, os.path.basename(args.data_path))
output_dir = os.path.join(args.output, 'filter')
output_dir_left = os.path.join(args.output, "left")
output_dir_right = os.path.join(args.output, "right")
if not args.no_cuda:
os.environ['CUDA_DEVICE_ORDER'] = "PCI_BUS_ID"
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu_id
use_cuda = torch.cuda.is_available()
match_files = GetImages(args.data_path)
if not match_files:
print("No matched files in the path: ", args.data_path)
exit(0)
left_files, right_files, root_len = match_files
model = Model(max_disp=256, mixed_precision=False, test_mode=True)
print('Number of model parameters: {}'.format(sum([p.data.nelement() for p in model.parameters()])))
ckpt = torch.load(args.model_file)
if 'optim_state_dict' in ckpt.keys():
model_state_dict = OrderedDict()
for k, v in ckpt['state_dict'].items():
name = k[7:] # remove `module.`
model_state_dict[name] = v
else:
model_state_dict = ckpt
model.load_state_dict(model_state_dict, strict=True)
if use_cuda:
model.cuda()
model.eval()
start = 0
left_files = left_files[start:]
right_files = right_files[start:]
for left_image_file, right_image_file in tzip(left_files, right_files):
if not os.path.exists(left_image_file) or not os.path.exists(right_image_file):
continue
output_name = left_image_file[root_len + 1:]
# todo: hao 2025-02-08 15:25 - update the logic
if args.ignore_target:
name = os.path.splitext(output_name)[0] + ".png"
output_concat_color = os.path.join(output_dir, "concat_color", name)
if os.path.exists(output_concat_color):
continue
left_img = cv2.imread(left_image_file)
right_img = cv2.imread(right_image_file)
org_h, org_w = left_img.shape[:2]
in_h, in_w = left_img.shape[:2]
scale_h, scale_w = [int(dim * args.scale) for dim in left_img.shape[:2]]
left_img = cv2.resize(left_img, (scale_w, scale_h), interpolation=cv2.INTER_LINEAR)
right_img = cv2.resize(right_img, (scale_w, scale_h), interpolation=cv2.INTER_LINEAR)
if in_h % 8 != 0:
pad_h = in_h % 8
left_img = np.pad(left_img, ((pad_h // 2, pad_h // 2), (0, 0), (0, 0)), mode='reflect')
right_img = np.pad(right_img, ((pad_h // 2, pad_h // 2), (0, 0), (0, 0)), mode='reflect')
if in_w % 8 != 0:
pad_w = in_w % 8
left_img = np.pad(left_img, ((0, 0), (pad_w // 2, pad_w // 2), (0, 0)), mode='reflect')
right_img = np.pad(right_img, ((0, 0), (pad_w // 2, pad_w // 2), (0, 0)), mode='reflect')
in_h, in_w = left_img.shape[:2]
if in_h % 8 != 0:
pad_h = in_h % 8
left_img = np.pad(left_img, ((pad_h // 2, pad_h // 2), (0, 0), (0, 0)), mode='reflect')
right_img = np.pad(right_img, ((pad_h // 2, pad_h // 2), (0, 0), (0, 0)), mode='reflect')
if in_w % 8 != 0:
pad_w = in_w % 8
left_img = np.pad(left_img, ((0, 0), (pad_w // 2, pad_w // 2), (0, 0)), mode='reflect')
right_img = np.pad(right_img, ((0, 0), (pad_w // 2, pad_w // 2), (0, 0)), mode='reflect')
eval_h, eval_w = left_img.shape[:2]
assert eval_h % 8 == 0, "input height should be divisible by 8"
assert eval_w % 8 == 0, "input width should be divisible by 8"
imgL = cv2.resize(left_img, (eval_w, eval_h), interpolation=cv2.INTER_LINEAR)
imgR = cv2.resize(right_img, (eval_w, eval_h), interpolation=cv2.INTER_LINEAR)
with torch.no_grad():
start = time()
disp_left = inference(imgL, imgR, model, n_iter=20)
print("inference use: {:.2f} ms".format((time() - start) * 1000))
left_flip = cv2.flip(left_img, 1)
right_flip = cv2.flip(right_img, 1)
dispR_filp = inference(right_flip, left_flip, model, n_iter=20)
disp_right = cv2.flip(dispR_filp, 1)
# clip depth,
# remove too large distance
# scale to origin size（if scale<1）
if abs(args.scale - 1) > 1e-6:
disp_left = cv2.resize(disp_left, (org_w, org_h), interpolation=cv2.INTER_LINEAR)
disp_right = cv2.resize(disp_right, (org_w, org_h), interpolation=cv2.INTER_LINEAR)
left_img_show = cv2.resize(left_img, (org_w, org_h), interpolation=cv2.INTER_LINEAR)
right_img_show = cv2.resize(right_img, (org_w, org_h), interpolation=cv2.INTER_LINEAR)
else:
left_img_show = left_img
right_img_show = right_img
WriteDepth(disp_left, left_img_show, output_dir_left, output_name, bf, args.max_depth)
WriteDepth(disp_right, right_img_show, output_dir_right, output_name, bf, args.max_depth)
if __name__ == '__main__':
main()