-
Notifications
You must be signed in to change notification settings - Fork 49
/
Copy pathinference.py
192 lines (174 loc) · 7.23 KB
/
inference.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
"""Inference demo of directional point detector."""
import math
import cv2 as cv
import numpy as np
import torch
from torchvision.transforms import ToTensor
import config
from data import get_predicted_points, pair_marking_points, calc_point_squre_dist, pass_through_third_point
from model import DirectionalPointDetector
from util import Timer
def plot_points(image, pred_points):
"""Plot marking points on the image."""
if not pred_points:
return
height = image.shape[0]
width = image.shape[1]
for confidence, marking_point in pred_points:
p0_x = width * marking_point.x - 0.5
p0_y = height * marking_point.y - 0.5
cos_val = math.cos(marking_point.direction)
sin_val = math.sin(marking_point.direction)
p1_x = p0_x + 50*cos_val
p1_y = p0_y + 50*sin_val
p2_x = p0_x - 50*sin_val
p2_y = p0_y + 50*cos_val
p3_x = p0_x + 50*sin_val
p3_y = p0_y - 50*cos_val
p0_x = int(round(p0_x))
p0_y = int(round(p0_y))
p1_x = int(round(p1_x))
p1_y = int(round(p1_y))
p2_x = int(round(p2_x))
p2_y = int(round(p2_y))
cv.line(image, (p0_x, p0_y), (p1_x, p1_y), (0, 0, 255), 2)
cv.putText(image, str(confidence), (p0_x, p0_y),
cv.FONT_HERSHEY_PLAIN, 1, (0, 0, 0))
if marking_point.shape > 0.5:
cv.line(image, (p0_x, p0_y), (p2_x, p2_y), (0, 0, 255), 2)
else:
p3_x = int(round(p3_x))
p3_y = int(round(p3_y))
cv.line(image, (p2_x, p2_y), (p3_x, p3_y), (0, 0, 255), 2)
def plot_slots(image, pred_points, slots):
"""Plot parking slots on the image."""
if not pred_points or not slots:
return
marking_points = list(list(zip(*pred_points))[1])
height = image.shape[0]
width = image.shape[1]
for slot in slots:
point_a = marking_points[slot[0]]
point_b = marking_points[slot[1]]
p0_x = width * point_a.x - 0.5
p0_y = height * point_a.y - 0.5
p1_x = width * point_b.x - 0.5
p1_y = height * point_b.y - 0.5
vec = np.array([p1_x - p0_x, p1_y - p0_y])
vec = vec / np.linalg.norm(vec)
distance = calc_point_squre_dist(point_a, point_b)
if config.VSLOT_MIN_DIST <= distance <= config.VSLOT_MAX_DIST:
separating_length = config.LONG_SEPARATOR_LENGTH
elif config.HSLOT_MIN_DIST <= distance <= config.HSLOT_MAX_DIST:
separating_length = config.SHORT_SEPARATOR_LENGTH
p2_x = p0_x + height * separating_length * vec[1]
p2_y = p0_y - width * separating_length * vec[0]
p3_x = p1_x + height * separating_length * vec[1]
p3_y = p1_y - width * separating_length * vec[0]
p0_x = int(round(p0_x))
p0_y = int(round(p0_y))
p1_x = int(round(p1_x))
p1_y = int(round(p1_y))
p2_x = int(round(p2_x))
p2_y = int(round(p2_y))
p3_x = int(round(p3_x))
p3_y = int(round(p3_y))
cv.line(image, (p0_x, p0_y), (p1_x, p1_y), (255, 0, 0), 2)
cv.line(image, (p0_x, p0_y), (p2_x, p2_y), (255, 0, 0), 2)
cv.line(image, (p1_x, p1_y), (p3_x, p3_y), (255, 0, 0), 2)
def preprocess_image(image):
"""Preprocess numpy image to torch tensor."""
if image.shape[0] != 512 or image.shape[1] != 512:
image = cv.resize(image, (512, 512))
return torch.unsqueeze(ToTensor()(image), 0)
def detect_marking_points(detector, image, thresh, device):
"""Given image read from opencv, return detected marking points."""
prediction = detector(preprocess_image(image).to(device))
return get_predicted_points(prediction[0], thresh)
def inference_slots(marking_points):
"""Inference slots based on marking points."""
num_detected = len(marking_points)
slots = []
for i in range(num_detected - 1):
for j in range(i + 1, num_detected):
point_i = marking_points[i]
point_j = marking_points[j]
# Step 1: length filtration.
distance = calc_point_squre_dist(point_i, point_j)
if not (config.VSLOT_MIN_DIST <= distance <= config.VSLOT_MAX_DIST
or config.HSLOT_MIN_DIST <= distance <= config.HSLOT_MAX_DIST):
continue
# Step 2: pass through filtration.
if pass_through_third_point(marking_points, i, j):
continue
result = pair_marking_points(point_i, point_j)
if result == 1:
slots.append((i, j))
elif result == -1:
slots.append((j, i))
return slots
def detect_video(detector, device, args):
"""Demo for detecting video."""
timer = Timer()
input_video = cv.VideoCapture(args.video)
frame_width = int(input_video.get(cv.CAP_PROP_FRAME_WIDTH))
frame_height = int(input_video.get(cv.CAP_PROP_FRAME_HEIGHT))
output_video = cv.VideoWriter()
if args.save:
output_video.open('record.avi', cv.VideoWriter_fourcc(*'XVID'),
input_video.get(cv.CAP_PROP_FPS),
(frame_width, frame_height), True)
frame = np.empty([frame_height, frame_width, 3], dtype=np.uint8)
while input_video.read(frame)[0]:
timer.tic()
pred_points = detect_marking_points(
detector, frame, args.thresh, device)
slots = None
if pred_points and args.inference_slot:
marking_points = list(list(zip(*pred_points))[1])
slots = inference_slots(marking_points)
timer.toc()
plot_points(frame, pred_points)
plot_slots(frame, pred_points, slots)
cv.imshow('demo', frame)
cv.waitKey(1)
if args.save:
output_video.write(frame)
print("Average time: ", timer.calc_average_time(), "s.")
input_video.release()
output_video.release()
def detect_image(detector, device, args):
"""Demo for detecting images."""
timer = Timer()
while True:
image_file = input('Enter image file path: ')
image = cv.imread(image_file)
timer.tic()
pred_points = detect_marking_points(
detector, image, args.thresh, device)
slots = None
if pred_points and args.inference_slot:
marking_points = list(list(zip(*pred_points))[1])
slots = inference_slots(marking_points)
timer.toc()
plot_points(image, pred_points)
plot_slots(image, pred_points, slots)
cv.imshow('demo', image)
cv.waitKey(1)
if args.save:
cv.imwrite('save.jpg', image, [int(cv.IMWRITE_JPEG_QUALITY), 100])
def inference_detector(args):
"""Inference demo of directional point detector."""
args.cuda = not args.disable_cuda and torch.cuda.is_available()
device = torch.device('cuda:' + str(args.gpu_id) if args.cuda else 'cpu')
torch.set_grad_enabled(False)
dp_detector = DirectionalPointDetector(
3, args.depth_factor, config.NUM_FEATURE_MAP_CHANNEL).to(device)
dp_detector.load_state_dict(torch.load(args.detector_weights))
dp_detector.eval()
if args.mode == "image":
detect_image(dp_detector, device, args)
elif args.mode == "video":
detect_video(dp_detector, device, args)
if __name__ == '__main__':
inference_detector(config.get_parser_for_inference().parse_args())