def main(argv):
if len(argv) < 1:
raise Exception(
"Please indicate the index of file you want to observe")
index = int(argv[0])
fileList = parse.listFileFromDir('meta')
data = np.load(fileList[index])
frameLabels = selectPedestrian(data, fast_forward=10)
print frameLabels
videoFileList = parse.listFileFromDir('video-data/processed')
video_filepath = videoFileList[index + 1]
tracking(video_filepath, frameLabels[0], frameLabels[1])
def main(argv):
# Argument Parsing
if len(argv) < 1 or int(argv[0]) > 5 or int(argv[0]) < 0:
printIndexInfo()
raise Exception("Please indicate the index of file you want to observe")
index = int(argv[0])
filepath = parse.listFileFromDir('meta')[index]
data = np.load(filepath)
refPt = crop_image(np.copy(data))
# print refPt
# Assume start at frame = 0
# feature, out_frame = simulateSample(data[0], refPt[0])
out_frame = simulateSample(data[0], refPt[0])
trackNextFrame(data[1], out_frame)
def main(argv):
# Argument Parsing
if len(argv) < 1 or int(argv[0]) > 5 or int(argv[0]) < 0:
printIndexInfo()
raise Exception(
"Please indicate the index of file you want to observe")
index = int(argv[0])
# List all files in the directory and load the data as specified
fileList = parse.listFileFromDir('meta')
data = np.load(fileList[index])
# Get the frame, top left corner, and top right corner from manual labelling
frame, refPt = selectPedestrian(np.copy(data), fast_forward=10)
# print refPt
# print keyPt
# Perform tracking
tracking(data, frame, refPt)
def main(argv):
# Argument Parsing
if len(argv) < 1 or int(argv[0]) > 5 or int(argv[0]) < 0:
printIndexInfo()
raise Exception(
"Please indicate the index of file you want to observe")
index = int(argv[0])
filepath = parse.listFileFromDir('meta')[index]
data = np.load(filepath)
refPt = crop_image(np.copy(data))
topLeft, bottomRight = np.array(refPt[0][0]), np.array(refPt[0][1])
imageWidth = abs(bottomRight[0] - topLeft[0])
imageHeight = abs(bottomRight[1] - topLeft[1])
extension = 15
numSample = 2000
cv2.namedWindow("NewFrame")
# 1. Construct Positive bags and Negative bags (All Positive are in one bag and Negative bag contain only one data)
positiveBags, negativeBags, positiveMaps, negativeMaps, labels = bagConstruction(
data[0],
topLeft,
bottomRight,
imageWidth,
imageHeight,
extension=extension,
numSample=numSample)
# print len(positiveBags[0]), len(negativeBags)
# 2. Train in MISVM
classifier.fit(positiveBags + negativeBags, labels)
# 3. For loop on each frame
for frame in xrange(1, data.shape[0], 3):
# 3.1 Crop a set of image using and compute feature vector based on MIL apperance model
bags, bagMaps = cropImage(data[frame],
np.array(topLeft),
np.array(bottomRight),
imageWidth,
imageHeight,
extension=extension,
numSample=numSample)
# 3.2 Use MI Classifier to estimate p(y=1|x) => apply MILBoost
predict = classifier.predict(bags)
maxPoint = bagMaps[predict.argsort()[-1]]
topLeft = tuple(
[int((maxPoint[0] - topLeft[0]) * 0.85 + topLeft[0]), topLeft[1]])
# topLeft = tuple([int(maxPoint[0] - imageWidth/2), topLeft[1]])#int(maxPoint[1] - imageHeight/2)])
bottomRight = tuple(
[int(maxPoint[0] + imageWidth / 2),
bottomRight[1]]) #int(maxPoint[1] + imageHeight/2)])
print "After update"
print topLeft, bottomRight
# 3.3 Update the location of the image
cv2.rectangle(data[frame], topLeft, bottomRight, (0, 255, 0), 1)
# 3.4 Crop out two images one for positive and another is negative
positiveBags, negativeBags, positiveMaps, negativeMaps, labels = bagConstruction(
data[frame],
np.array(topLeft),
np.array(bottomRight),
imageWidth,
imageHeight,
extension=extension,
numSample=numSample)
# 3.5 Update MIL appearance model
classifier.fit(positiveBags + negativeBags, labels)
while True:
cv2.imshow("NewFrame", data[frame])
key = cv2.waitKey(1) & 0xFF
if key == ord("n"):
break
cv2.destroyAllWindows()