def loadrecords(csvfile, cachedir):
'''
Loads the records described by the csv file and cropped images in the
cachedir.
This returns xdata as a numpy 3D array, which is an array of 2D images.
The ydata will be a numpy 1D array of values of 1 or -1. The ydata array
represents the sex field of the csv file. +1 represents male and -1
represents female.
>>> xdata, ydata = loadrecords(csvfile, cachedir)
'''
xdata = []
ydata = []
with open(csvfile, 'r') as trainFile:
reader = csv.DictReader(trainFile, dialect='excel-tab')
for line in reader:
filename = os.path.basename(line['imagePath'])
split = os.path.splitext(filename)
cachepic = split[0] + '-' + line['line'] + '.png'
xdata.append(imagefuncs.loadImage(os.path.join(cachedir, cachepic)))
ydata.append(1 if line['truth-sex'] == 'M' else -1)
return np.asarray(xdata), np.asarray(ydata)
def main(arguments):
'Main entry point'
args = parseArgs(arguments)
with open(args.classifier, 'rb') as classifierFile:
classifier = cPickle.load(classifierFile)
xdata = []
imagepaths = []
print 'Reading in images...'
for entry in os.listdir(args.input):
filepath = os.path.join(args.input, entry)
if os.path.isfile(filepath):
try:
xdata.append(loadImage(filepath))
except IOError:
# This probably wasn't an image file or bad permissions
print >> sys.stderr, 'Warning: Could not read image', filepath
else:
# Only store the filepath if we were able to load the image
imagepaths.append(filepath)
print ' done'
print 'Preprocessing data...'
xdata = np.asarray(xdata)
xdata = preprocessData(xdata, reshape=(not isinstance(classifier, ConvNet)))
print ' done'
print 'Classifying the data...'
labels = classifier.predict(xdata)
print ' done'
# Output the results to a file
with open(args.output, 'w') as outfile:
writer = csv.writer(outfile, dialect='excel-tab')
writer.writerow(['imagepath', 'classification'])
for imagepath, label in zip(imagepaths, labels):
labelString = 'M' if label == 1 else 'F'
writer.writerow([imagepath, labelString])
print 'Wrote classification data to', args.output
