def __init__(self, path):
"""You provide the path of the dataset and then the input to channel 0 is the background subtraction mask."""
self.path = path
# Open the region of interest and convert it into a mask we can use...
roi = cv.LoadImage(os.path.join(path, 'ROI.bmp'))
self.roi = (cv2array(roi)!=0).astype(numpy.uint8)
if len(self.roi.shape)==3: self.roi = self.roi[:,:,0]
# Get the temporal range of the frames we are to analyse...
data = open(os.path.join(path, 'temporalROI.txt'), 'r').read()
tokens = data.split()
assert(len(tokens)==2)
self.start = int(tokens[0])
self.end = int(tokens[1]) # Note that these are inclusive.
# Confusion matrix - first index is truth, second guess, bg=0, fg=1...
self.con = numpy.zeros((2,2), dtype=numpy.int32)
# Shadow matrix - for all pixels that are in shadow records how many are background (0) and how many are foreground (1)...
self.shadow = numpy.zeros(2, dtype=numpy.int32)
# Basic stuff...
self.frame = 0
self.video = None
self.channel = 0
def __init__(self, path):
"""You provide the path of the dataset and then the input to channel 0 is the background subtraction mask."""
self.path = path
# Open the region of interest and convert it into a mask we can use...
roi = cv.LoadImage(os.path.join(path, 'ROI.bmp'))
self.roi = (cv2array(roi)!=0).astype(numpy.uint8)
if len(self.roi.shape)==3: self.roi = self.roi[:,:,0]
# Get the temporal range of the frames we are to analyse...
data = open(os.path.join(path, 'temporalROI.txt'), 'r').read()
tokens = data.split()
assert(len(tokens)==2)
self.start = int(tokens[0])
self.end = int(tokens[1]) # Note that these are inclusive.
# Confusion matrix - first index is truth, second guess, bg=0, fg=1...
self.con = numpy.zeros((2,2), dtype=numpy.int32)
# Shadow matrix - for all pixels that are in shadow records how many are background (0) and how many are foreground (1)...
self.shadow = numpy.zeros(2, dtype=numpy.int32)
# Basic stuff...
self.frame = 0
self.video = None
self.channel = 0
def nextFrame(self):
self.frame = cv.QueryFrame(self.vid)
if self.frame==None: return False
self.frameNP = cv2array(self.frame)[:,:,::-1].astype(numpy.float32)/255.0
return True
def nextFrame(self):
self.frame = cv.QueryFrame(self.vid)
if self.frame == None: return False
self.frameNP = cv2array(self.frame)[:, :, ::-1].astype(
numpy.float32) / 255.0
return True
def nextFrame(self):
if self.index<len(self.files):
try:
img = cv.LoadImage(self.files[self.index])
#print img.nChannels, img.width, img.height, img.depth, img.origin
self.frame = cv2array(img)[:,:,::-1].astype(numpy.float32)/255.0
except:
print 'Frame #%i with filename %s failed to load.'%(self.index,self.files[self.index])
self.index += 1
else:
self.frame = None
return self.frame!=None
def nextFrame(self):
if self.index < len(self.files):
try:
img = cv.LoadImage(self.files[self.index])
#print img.nChannels, img.width, img.height, img.depth, img.origin
self.frame = cv2array(img)[:, :, ::-1].astype(
numpy.float32) / 255.0
except:
print 'Frame #%i with filename %s failed to load.' % (
self.index, self.files[self.index])
self.index += 1
else:
self.frame = None
return self.frame != None
def nextFrame(self):
# Increase frame number - need to be 1 based for this...
self.frame += 1
if self.frame>=self.start and self.frame<=self.end:
# Fetch the provided mask...
mask = self.video.fetch(self.channel)
# Load the ground truth file...
fn = os.path.join(self.path, 'groundtruth/gt%06i.png'%self.frame)
gt = cv2array(cv.LoadImage(fn))
gt = gt.astype(numpy.uint8)
if len(gt.shape)==3: gt = gt[:,:,0]
# Loop the pixels and analyse each one, summing into the confusion matrix...
code = start_cpp() + """
for (int y=0; y<Ngt[0]; y++)
{
for (int x=0; x<Ngt[1]; x++)
{
if ((ROI2(y,x)!=0)&&(GT2(y,x)!=170))
{
int t = (GT2(y,x)==255)?1:0;
int g = (MASK2(y,x)!=0)?1:0;
CON2(t,g) += 1;
if (GT2(y,x)==50)
{
SHADOW1(g) += 1;
}
}
}
}
"""
roi = self.roi
con = self.con
shadow = self.shadow
weave.inline(code, ['mask', 'gt', 'roi', 'con', 'shadow'])
return self.frame<=self.end
def nextFrame(self):
# Increase frame number - need to be 1 based for this...
self.frame += 1
if self.frame>=self.start and self.frame<=self.end:
# Fetch the provided mask...
mask = self.video.fetch(self.channel)
# Load the ground truth file...
fn = os.path.join(self.path, 'groundtruth/gt%06i.png'%self.frame)
gt = cv2array(cv.LoadImage(fn))
gt = gt.astype(numpy.uint8)
if len(gt.shape)==3: gt = gt[:,:,0]
# Loop the pixels and analyse each one, summing into the confusion matrix...
code = start_cpp() + """
for (int y=0; y<Ngt[0]; y++)
{
for (int x=0; x<Ngt[1]; x++)
{
if ((ROI2(y,x)!=0)&&(GT2(y,x)!=170))
{
int t = (GT2(y,x)==255)?1:0;
int g = (MASK2(y,x)!=0)?1:0;
CON2(t,g) += 1;
if (GT2(y,x)==50)
{
SHADOW1(g) += 1;
}
}
}
}
"""
roi = self.roi
con = self.con
shadow = self.shadow
weave.inline(code, ['mask', 'gt', 'roi', 'con', 'shadow'])
return self.frame<=self.end
