-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
executable file
·201 lines (164 loc) · 8.27 KB
/
main.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
#coding=utf-8
from __future__ import print_function
import time
import argparse
from glob import glob
import os, cv2
import numpy as np
import matplotlib.pyplot as plt
from detector.base_detector import BaseDetector
# from detector.tf_faster_rcnn0413.tools.tf_faster_rcnn_detector import TfFasterRcnnDetector
# from detector.tf_faster_rcnn0413.tools.tf_faster_rcnn_mcdc_detector import TfFasterRcnnDetector
from detector.tf_faster_rcnn0413.tools.tf_faster_rcnn_coco_detector import TfFasterRcnnDetector
#from detector.darknet.yolov3_detector import YoloV3Detector
from predictor.base_predictor import BasePredictor
from predictor.test_predictor import TestBasePredictor
from predictor.monodepth.monodepth_simple_for_video import MonodepthPredictor
from predictor.lstm_predictor_x_vx import LSTMPredictor
from predictor.car_selector import CarSelector
#from predictor.area_predictor import AreaPredictor
from predictor.regression_predictor import RegressionPredictor
from tools.video_reader import VideoReader
from tools.choice_bbox import find_middle_car_use_iou, vis_detections
NETS = {'vgg16': ('vgg16_faster_rcnn_iter_70000.ckpt',),
'res101': ('res101_faster_rcnn_iter_110000.ckpt','res101_faster_rcnn_iter_1190000.ckpt')}
DATASETS= {'pascal_voc': ('voc_2007_trainval',),'pascal_voc_0712': ('voc_2007_trainval+voc_2012_trainval',),
'coco':('coco_2014_train+coco_2014_valminusminival',),
'mcdc':('mcdc_train_10000',)
}
PREDICTOR = {'citys': 'model_cityscapes', 'kitti' : 'model_kitti', 'city2kitti':'model_city2kitti',}
def read_data(video, time_file):
vreader = VideoReader(video)
times = []
with open(time_file, 'r') as fin:
for line in fin.readlines():
times.append(float(line))
print('Total frames: ', len(times), times[0], times[-1])
return vreader, times
def parse_args():
parser = argparse.ArgumentParser(description='Demo for MCDC, copyright@Ready Player One')
parser.add_argument('--input-dir', required=True,
help='directory for valid or test video', type=str)
parser.add_argument('--output-dir', required=True,
help='directory for result', type=str)
parser.add_argument('-c', '--cam-calib', required=True,
help='calibrated camera parameter file path')
#for depth predictor
parser.add_argument('--predictor_model', dest='predictor_model', help='Model to use [citys kitti]',
choices=PREDICTOR.keys(), default='city2kitti')
# parser.add_argument('--image_path', type=str, help='path to the image', required=True)
parser.add_argument('--encoder', type=str, help='type of encoder, vgg or resnet50', default='vgg')
parser.add_argument('--input_height', type=int, help='input height', default=256)
parser.add_argument('--input_width', type=int, help='input width', default=512)
parser.add_argument('--detector', dest='detector',
help='tf-faster-rcnn', default='tf-faster-rcnn', type=str)
parser.add_argument('--detector_net', dest='detector_net', help='Network to use [vgg16 res101]',
choices=NETS.keys(), default='res101')
parser.add_argument('--dataset', dest='dataset', help='Trained dataset [pascal_voc pascal_voc_0712 coco mcdc]',
choices=DATASETS.keys(), default='coco')
parser.add_argument('--x_vx_mode', type = str, help="mode for lstm predictor ['x','vx', 'x_vx']",
default='x_vx')
parser.add_argument('--learning_rate', type=float, help='learning_rate', default=0.006)
parser.add_argument('--lstm_predictor_model',
help='directory for lstm predictor models', type=str, default = 'models/')
parser.add_argument('--data-type', help='val, test', type=str, default='test')
parser.add_argument('--gpu', type=int, default=0,
help='choose one gpu for distribution computing')
args = parser.parse_args()
return args
def save_to_img(video, img, bbox, fid, depth_img=None, depth_bbox=None, pre=None):
show_dir = 'show/%s' % os.path.basename(video)
cv2.rectangle(img,
(int(bbox[0]), int(bbox[1])),
(int(bbox[2]), int(bbox[3])),
(0, 255, 0), 3
)
x = int((bbox[0] + bbox[2])/2)
y = int(bbox[3])
cv2.rectangle(img,
(x, y),
(x, y),
(0, 0, 255), 10
)
if pre is not None:
font = cv2.FONT_HERSHEY_SIMPLEX
img = cv2.putText(img, 'Relative distance: %f m'%(pre['x']), (50, 50), font, 1.2, (255, 0, 0), 2)
img = cv2.putText(img, 'Relative speed : %f m/s' % (pre['vx']), (50, 100), font, 1.2, (255, 0, 0), 2)
if not os.path.exists(show_dir):
os.makedirs(show_dir)
cv2.imwrite('%s/single_bbox_%d.jpg' % (show_dir, fid), img)
if depth_img is not None:
# cv2.imwrite('%s/%d_depth.png' % (show_dir, fid), depth_img)
# cv2.imwrite('%s/%d_depth_bbox.png' % (show_dir, fid), depth_bbox)
# depth_to_img = scipy.misc.imresize(depth_img, [256, 512])
plt.imsave('%s/%d_depth.png' % (show_dir, fid), depth_img, cmap='plasma')
# depth_bbox_to_img = scipy.misc.imresize(depth_bbox, [300, 400])
if depth_bbox is not None:
plt.imsave('%s/%d_depth_bbox.png' % (show_dir, fid), depth_bbox, cmap='plasma')
# plt.imsave(os.path.join(output_directory, "{}_disp.png".format(output_name)), disp_to_img, cmap='plasma')
def main():
args = parse_args()
print(args)
# 1. browse videos and time file in input dir
videos = glob(os.path.relpath(args.input_dir) + '/*video*.avi')
time_files = [v[:-4] + '_time.txt' for v in videos] # second
gt_files = [v[:-4] + '_gt.json' for v in videos]
# 2. init detector model
if args.detector == 'tf-faster-rcnn':
detector = TfFasterRcnnDetector(args)
# elif args.detector == 'yolo-v3':
# detector = YoloV3Detector(args)
else:
detector = BaseDetector(args)
selector = CarSelector(args)
# 3. process video one by one
for video, time_file, gt_file in zip(videos, time_files, gt_files):
# if int(video[-6:-4]) % 6 != args.gpu - 2:
# continue
# if int(video[-6:-4]) < 12:
# continue
print(video, time_file)
# predictor = BasePredictor(args)
# predictor = RegressionPredictor(args)
predictor = MonodepthPredictor(args)
# predictor = LSTMPredictor(args)
vreader, times = read_data(video, time_file)
before = 500
start = time.time()
last_bbox = np.zeros((1, 5), dtype=np.float32)
for fid, t in enumerate(times):
# print('fid: ', fid)
frame = vreader.next()
dets = detector.detect(frame)
if fid == 0:
# bbox = find_middle_car_use_iou(dets, rang=np.array([600, 500, 1160, 900]))
bbox = selector.select_front_car(dets)
else:
bbox = selector.find_middle_car_use_iou(dets, rang=last_bbox)
last_bbox = bbox
#for MonodepthPredictor
pre = predictor.predict(bbox, t, fid, frame)
cur_pred, depth_img, depth_img_bbox = pre
#for others
# pre = predictor.predict(bbox, t, fid, frame)
if fid % 1 == 0:
save_to_img(video, frame, bbox, fid, depth_img, depth_img_bbox, cur_pred)
vis_detections(frame, 'car', dets, thresh=0.5, video=video, fid=fid)
if fid >= before:
break
print('>> Time elapsed: %lf s' % (time.time() - start))
if not os.path.exists(args.output_dir):
os.makedirs(args.output_dir)
print('makedirs ', args.output_dir)
result_file = os.path.join(args.output_dir, os.path.basename(video)[:-4] + '_pre.json')
predictor.to_json(result_file)
# visual for debug
if args.data_type == 'test':
predictor.draw_test_line(video)
pass
else:
predictor.draw_valid_line(gt_file, fid)
predictor.err_estimation(gt_file, fid)
if __name__ == '__main__':
main()
# CUDA_VISIBLE_DEVICES=0 python main.py --input-dir=./test --output-dir=./test_pre --cam-calib=./test/camera_parameter.json