-
Notifications
You must be signed in to change notification settings - Fork 1
/
process_videos.py
321 lines (272 loc) · 12 KB
/
process_videos.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
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
import numpy as np
import pandas as pd
import imutils
import time
import cv2
import os
from sklearn.metrics import matthews_corrcoef
#################### UTILITY FUNCTIONS ####################
def blend_images(frame_list):
for idx, img in enumerate(frame_list,1):
if idx == 1:
first_img = img
continue
else:
second_img = img
second_weight = 1/(idx+1)
first_weight = 1 - second_weight
first_img = cv2.addWeighted(first_img, first_weight, second_img, second_weight, 0)
return first_img
def getOrientation(pts):
sz = len(pts)
data_pts = np.empty((sz, 2), dtype=np.float64)
for i in range(data_pts.shape[0]):
data_pts[i,0] = pts[i,0,0]
data_pts[i,1] = pts[i,0,1]
# Perform PCA analysis
mean = np.empty((0))
mean, eigenvectors, eigenvalues = cv2.PCACompute2(data_pts, mean)
angle = cv2.phase(eigenvectors[1,1], eigenvectors[1,0],angleInDegrees=True)
# convert angle to 0-180
angle180 = angle[0][0] if angle[0][0]<180 else angle[0][0]-180
return angle180
def orientationDiff(angle1, angle2):
diff = abs(angle1 - angle2)
diff90 = diff if diff<90 else 180-diff
return diff90
def write_image(directory, output_folder, image_name, image):
if not os.path.exists(os.path.join(directory,output_folder)):
os.mkdir(os.path.join(directory,output_folder))
cv2.imwrite(os.path.join(directory, output_folder, clean_name+'.jpg'), image )
def read_video(video_path):
# start the video stream thread
cap = cv2.VideoCapture(video_path)
time.sleep(1.0)
(grabbed, frame) = cap.read()
# find orange border
gaussian = cv2.GaussianBlur(frame, (3, 3), 0)
frame_HSV = cv2.cvtColor(gaussian, cv2.COLOR_BGR2HSV)
orange_border = cv2.inRange(frame_HSV, (0, 100, 100), (50, 255, 255))
# make border a bit bolder
kernel = np.ones((3,3),np.uint8)
dilation = cv2.dilate(orange_border,kernel,iterations = 1)
# make border grey
grey_border = cv2.threshold(dilation, 100, 150, cv2.THRESH_BINARY)[1]
# find contours (i.e., outlines) of the foreground objects in the
# thresholded image
cnts = cv2.findContours(grey_border.copy(), cv2.RETR_LIST,
cv2.CHAIN_APPROX_NONE)
cnts = imutils.grab_contours(cnts)
src = grey_border.copy()
# find true contour (inside of mask)
if len(cnts) <= 1:
return np.zeros((384, 512, 1), np.uint8), 0, 0
cnts_df = pd.DataFrame([cv2.contourArea(c) for c in cnts], columns = ['area'])
idx = cnts_df.sort_values(by=['area'], ascending=False).iloc[[1]].index[0]
c=cnts[idx]
cv2.drawContours(src, [c], -1, (255, 255, 255), -1)
mask = cv2.threshold(src, 200, 255, cv2.THRESH_BINARY)[1]
#---------------PCA-mask--------------------
contours, _ = cv2.findContours(mask, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)
c = contours[0]
# Calculate the angle of mask
angle_mask = getOrientation(c)
# Calculate area of mask
area_mask = cv2.contourArea(c)
#--------------------------------------------
# read frames and add to a list
frame_list= []
while True:
# if this is a file video stream, then we need to check if
# there any more frames left in the buffer to process
(grabbed, frame) = cap.read()
if not grabbed:
break
# grab the frame from the threaded video file stream, resize
# it, and convert it to grayscale
# channels)
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
output = cv2.bitwise_and(gray, gray, mask=mask)
frame_list.append(output)
return frame_list, angle_mask, area_mask
#################### SIMPLE BLENDING METHOD ####################
# just blend all frames
def simple_blending_method(frame_list):
blended = blend_images(frame_list)
# simple threshold = 65
sharpened = cv2.threshold(blended, 65, 255, cv2.THRESH_BINARY)[1]
# otsu
blur = cv2.GaussianBlur(sharpened,(5,5),0)
th_otsu, otsu = cv2.threshold(blur,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)
# closing
kernel=np.ones((3,3),np.uint8)
closing = cv2.morphologyEx(otsu, cv2.MORPH_CLOSE, kernel)
# opening
kernel=np.ones((2,2),np.uint8)
opening = cv2.morphologyEx(closing, cv2.MORPH_OPEN, kernel)
#remove little contours
eliminated_contours = []
contours, _ = cv2.findContours(opening, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)
for c in contours:
# Calculate the area of each contour
area = cv2.contourArea(c)
# Filter contours that are too small or too large
if area < 1e2 or 1e5 < area:
eliminated_contours.append(c)
# fill black unfiltered contours
cv2.drawContours(opening, eliminated_contours, -1, (0, 0, 0), -1)
return opening
#################### CONTOUR ORIENTATITON METHOD 1: blending contours ####################
# construct short flows by blending frames with a window, check contour orientations in flows,
# if they match with orientation of mask, blend contours
def contour_orientation_method1(frame_list, angle_mask, area_mask):
filtered_flows=[]
window = 5
for i in range(len(frame_list)-window-1):
flow = blend_images(frame_list[i : i+window])
# median blur
median = cv2.medianBlur(flow,7)
# threshold = 65
sharpened = cv2.threshold(median, 65, 255, cv2.THRESH_BINARY)[1]
# otsu
blur = cv2.GaussianBlur(sharpened,(5,5),0)
th,otsu = cv2.threshold(blur,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)
#--------------PCA of contours -----------
eliminated_contours = []
contours, _ = cv2.findContours(otsu, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)
if len(contours)==0:
continue
for c in contours:
# Calculate the area of each contour
area = cv2.contourArea(c)
# Filter contours that are too small or too large
if area < area_mask*0.01 or area_mask <= area:
eliminated_contours.append(c)
continue
angle = getOrientation(c)
# Filter contours which have different orientation than mask
if (orientationDiff(angle,angle_mask) > 45):
eliminated_contours.append(c)
#------------------------------------------
# fill black eliminated contours
cv2.drawContours(otsu, eliminated_contours, -1, (0, 0, 0), -1)
# add result to list
filtered_flows.append(otsu)
# comment here to see errors
if len(filtered_flows) == 0:
return np.zeros((384, 512, 1), np.uint8)
# blend filtered frames
blended_flows = blend_images(filtered_flows)
# otsu
gaus = cv2.GaussianBlur(blended_flows,(5,5),0)
th_otsu,otsu = cv2.threshold(gaus,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)
#remove little contours
eliminated_contours = []
contours, _ = cv2.findContours(otsu, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)
for c in contours:
# Calculate the area of each contour
area = cv2.contourArea(c)
# Filter contours that are too small or too large
if area < area_mask*0.01 or area_mask <= area:
eliminated_contours.append(c)
# fill black eliminated contours
cv2.drawContours(otsu, eliminated_contours, -1, (0, 0, 0), -1)
return otsu
#################### CONTOUR ORIENTATITON METHOD 2: blending frames ####################
# construct short flows by blending frames with a window, check orientation of biggest contour in flows,
# if it match with orientation of mask, blend flows
def contour_orientation_method2(frame_list, angle_mask, area_mask):
filtered_frames=[]
window = 10
for i in range(len(frame_list)-window-1):
flow = blend_images(frame_list[i : i+window])
# clahe = cv2.createCLAHE(clipLimit=65, tileGridSize=(8,8))
# c1 = clahe.apply(flow)
# threshold = 65
sharpened = cv2.threshold(flow, 65, 255, cv2.THRESH_BINARY)[1]
# otsu
blur = cv2.GaussianBlur(sharpened,(5,5),0)
th,otsu = cv2.threshold(blur,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)
#--------------PCA of biggest contour -----------
contours, _ = cv2.findContours(otsu, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)
if len(contours)==0:
continue
area_list=[]
for c in contours:
# Calculate the area of each contour
area = cv2.contourArea(c)
area_list.append(area)
# Check if biggest contour has different orientation than mask
max_idx = np.argmax(area_list)
angle = getOrientation(contours[max_idx])
if (orientationDiff(angle,angle_mask) < 45):
filtered_frames.append(flow)
#------------------------------------------
# comment here to see errors
if len(filtered_frames) == 0:
return np.zeros((384, 512, 1), np.uint8)
# blend filtered frames
blended_flows = blend_images(filtered_frames)
# threshold = 65
basicth = cv2.threshold(blended_flows, 65, 255, cv2.THRESH_BINARY)[1]
# otsu
gaus = cv2.GaussianBlur(basicth,(5,5),0)
th_otsu,otsu = cv2.threshold(gaus,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)
#remove little contours
eliminated_contours = []
contours, _ = cv2.findContours(otsu, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)
for c in contours:
# Calculate the area of each contour
area = cv2.contourArea(c)
# Filter contours that are too small or too large
if area < area_mask*0.01 or area_mask <= area :
eliminated_contours.append(c)
# fill black unfiltered contours
cv2.drawContours(otsu, eliminated_contours, -1, (0, 0, 0), -1)
return otsu
#################### MAIN ####################
directory = "./micro"
output_folder = "method1-nomedian"
files = os.listdir(directory)
videos = [file for file in files if os.path.splitext(file)[1] == '.mp4']
# read data
metadata_df = pd.read_csv("./metadata/train_metadata.csv", sep=',')
labels_df = pd.read_csv("./metadata/train_labels.csv", sep=',')
#tier1_micro = [video for video in videos if video in metadata_df.loc[metadata_df['tier1']==True, 'filename'].to_list()]
#tier1_micro_stalled = [video for video in tier1_micro if video in labels_df.loc[labels_df['stalled']==1, 'filename'].to_list()]
predictions = []
contour_counts=[]
for i, video_name in enumerate(videos):
print("[INFO] processing index: ", i, " video name: ", video_name, " is started.")
clean_name = os.path.splitext(video_name)[0]
file_path = os.path.join(directory,video_name)
frame_list, angle_mask, area_mask = read_video(file_path)
image = contour_orientation_method1(frame_list, angle_mask, area_mask)
write_image(directory, output_folder, clean_name, image)
print(" ... done.")
# predictions
contours, _ = cv2.findContours(image, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)
contour_counts.append(len(contours))
if len(contours)>1:
predictions.append(1)
else:
predictions.append(0)
# evaluation
true_labels_micro = [labels_df.loc[labels_df['filename']==video, 'stalled'].squeeze() for video in videos]
pred_score_micro = matthews_corrcoef(true_labels_micro, predictions)
print(pred_score_micro)
# export outputs
predictions_df = pd.DataFrame(zip(contour_counts,predictions), columns = ["contour_counts","predictions"])
predictions_df.to_csv("predictions_method1-nomedian.csv", encoding='utf-8')
# 0.23694613809322196 ->contour_orientation_method1 no median
# 0.22344012669581176 -> contour_orientation_method1 median
# 0.2662261064410542 ->contour_orientation_method2
# 0.17265817672643158 ->simple
# 0.09691587345338028 -> cnn-basic contour_orientation_method1 no median
# 0.10080577717127491 -> cnn complex - dropout(0.5)
# 0.1158313820809831 -> cnn complex - no dropout 0.12617454075567316
# 0.1648599173123011 -> cnn complex - no dense
# 0.06664070419513618 -> epoch 30
# 0.05488729658746616 -> cnn method2 no dropout epoch 3
# 0.1272300121627428 -> cnn method1 median dropout epoch 3
# 0.02491895841036883 -> cnn method1 median dropout epoch 3 no dense