-
Notifications
You must be signed in to change notification settings - Fork 0
/
custom_cov_yolov4_video_nms.py
196 lines (158 loc) · 8.35 KB
/
custom_cov_yolov4_video_nms.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
"""Created on Wed Sep 23 21:46:20 2020
@author: vibhanshusindhu"""
""" custom_cov_yolov4_video_nms """
import numpy as np
import cv2
# get the video stream
file_video_stream = cv2.VideoCapture('images/testing/cov_video.mp4')
# create a while loop # loop till the last frame of the video
while (file_video_stream.isOpened()):
# for every iteration obtain the current frame from video stream by using
# the read function
ret, current_frame = file_video_stream.read()
# ret is a bool value which indicates whether the read function was able
# to read the video stream correctly or not
# pass this current_frame to the object detector instead of testing image
img_to_detect = current_frame
img_height = img_to_detect.shape[0]
img_width = img_to_detect.shape[1]
# convert the image to blob(binary large object) to pass into model
img_blob = cv2.dnn.blobFromImage(img_to_detect, 0.003922, (416,416),
swapRB = True, crop = False)
# recommended by yolo authors, scale factor is 0.003922 = 1/255 (255 is the
# maximum color value of a pixel in the image), width,height of blob is
# (320,320). Accepted sizes are (320 x 320), (416 x 416), (609 x 609).More
# size means more accuracy but less speed. swapRB is set true bcoz originally
# the colour format in the image is RGB(red,green,blue) but opencv library
# converts it to BGR(blue,green,red), so to use it as binary large image we
# convert it back to RGB.
# only single label
class_labels = ["coronavirus"]
# declare only a single color
# only one color green since only one object to detect
# Split based on ',' and for every split, change type to int
# convert that to a numpy array to apply color mask to the image numpy array
class_colors = ["0,255,0"]
# print(class_colors)
class_colors = [np.array(every_color.split(",")).astype(int)
for every_color in class_colors]
# print(class_colors)
class_colors = np.array(class_colors)
# print(class_colors)
class_colors = np.tile(class_colors, (1,1))
# print(class_colors)
# loading the coronavirus custom model
yolo_model = cv2.dnn.readNetFromDarknet('model/cov_yolov4.cfg',
'model/cov_yolov4_best.weights')
# Get all layers from the yolo network
# Loop & find the last layer(output layer) of the yolo network
yolo_layers = yolo_model.getLayerNames()
# print(yolo_layers)
yolo_output_layer = [yolo_layers[yolo_layer[0] - 1] for yolo_layer in
yolo_model.getUnconnectedOutLayers()]
# print(yolo_output_layer)
# input preprocessed blob into the model & pass through the model
yolo_model.setInput(img_blob)
# Obtain the detection predictions by the model using forward() method or
# obtain the detection layers by forwarding through till the output layer
obj_detection_layers = yolo_model.forward(yolo_output_layer)
# print(obj_detection_layers)
######## NMS Change 1 ########
# initialization for non-maximum supression(NMS)
# declare list for [class id], [box_centre, width & height], [confidences]
class_ids_list = []
boxes_list = []
confidences_list = []
######## NMS Change 1 END ########
# loop over each of the layer outputs
for object_detection_layer in obj_detection_layers:
# loop over the detections
for object_detection in object_detection_layer:
# object_detection[1 to 4] => will have two centre points, box width
# & box height
# obj_detection[5] => will have scores for all objects within
# bounding box
all_scores = object_detection[5:]
predicted_class_id = np.argmax(all_scores)
prediction_confidence = all_scores[predicted_class_id]
# take only predictions with confidence more than 20%
if prediction_confidence > 0.20:
# get the predicted label
predicted_class_label = class_labels[predicted_class_id]
# obtain the bounding box co-oridnates for actual image from
# resized image size
bounding_box = object_detection[0:4] * np.array([img_width,
img_height,
img_width,
img_height])
(box_centre_x_pt, box_centre_y_pt, box_width, box_height
) = bounding_box.astype(int)
start_x_pt = int(box_centre_x_pt - (box_width/2))
start_y_pt = int(box_centre_y_pt - (box_height/2))
############## NMS Change 2 ###############
# save class id, start x, y, width & height, confidences in a list
# for nms processing
# make sure to pass confidence as float and width and height as
# integers
class_ids_list.append(predicted_class_id)
confidences_list.append(float(prediction_confidence))
boxes_list.append([start_x_pt, start_y_pt, int(box_width),
int(box_height)])
############## NMS Change 2 END ###########
############## NMS Change 3 ###############
# Applying the NMS will return only the selected max value ids while
# suppressing the non maximum (weak) overlapping bounding boxes
# Non-Maxima Suppression confidence set as 0.5 & max_suppression threshold
# for NMS as 0.4 (adjust and try for better perfomance)
max_value_ids = cv2.dnn.NMSBoxes(boxes_list, confidences_list, 0.5,0.4)
# max_value_ids will be a list of class_ids sorted descendingly in the order
# of their confidence_values
# loop through the final set of detections remaining after NMS and draw
# bounding box and write text
my_dict = {} # to keep count of numbers of same class object eg.cars
for max_valueid in max_value_ids:
max_class_id = max_valueid[0]
box = boxes_list[max_class_id]
start_x_pt = box[0]
start_y_pt = box[1]
box_width = box[2]
box_height = box[3]
#get the predicted class id and label
predicted_class_id = class_ids_list[max_class_id]
predicted_class_label = class_labels[predicted_class_id]
prediction_confidence = confidences_list[max_class_id]
############## NMS Change 3 END ###########
end_x_pt = start_x_pt + box_width
end_y_pt = start_y_pt + box_height
# get a random mask color from the numpy array of colors
box_color = class_colors[predicted_class_id]
# print(box_color)
# convert the color numpy array as a list and apply to text & box
box_color = [int(c) for c in box_color]
#print(box_color)
count = 1
#print(my_dict.keys())
if predicted_class_label in my_dict.keys():
count = my_dict[predicted_class_label]
tmp = predicted_class_label
# print the prediction in console
s = f"{predicted_class_label} {count} : {(prediction_confidence*100):.2f}"
predicted_class_label = s
print(f"predicted object {predicted_class_label}")
count+= 1
my_dict[tmp] = count
# draw rectangle and text in the image
cv2.rectangle(img_to_detect, (start_x_pt, start_y_pt),
(end_x_pt, end_y_pt), box_color, 1)
cv2.putText(img_to_detect, predicted_class_label,
(start_x_pt, start_y_pt-5), cv2.FONT_HERSHEY_SIMPLEX,
0.5, box_color, 1)
cv2.imshow("Detection Output", img_to_detect)
# terminate while loop if 'q' is pressed
if cv2.waitKey(1) & 0xFF == ord('q'):
break
# if keypress detected, release the stream & camera
# releasing the stream & camera
# then destroy/close all opencv windows
file_video_stream.release()
cv2.destroyAllWindows()