def LoadDatasetBaidu(filename):
file_list = np.genfromtxt(filename,
delimiter=',',
names=True,
dtype=np.dtype([('image', object),
('label', int), ('x', float),
('y', float), ('z', float),
('rx', float), ('ry', float),
('rz', float),
('timestamp', object)]))
input_images = ImageProcProxy.emptyDataset(file_list.shape[0],
params['input']['width'] * 2,
params['input']['height'])
orig_image_file = ImageProcProxy.readImageColor(file_list['image'][0])
image_crop = cv2.resize(
orig_image_file, (params['input']['width'], params['input']['height']))
image_gaus = ImageProcProxy.applyGaussian(
image_crop, params['filter']['gaussian_radius'],
params['filter']['gaussian_sigma'])
image_crop_int = ImageProcProxy.convertBGR2INT(image_crop)
image_gaus_int = ImageProcProxy.convertBGR2INT(image_gaus)
image_crop_vec = ImageProcProxy.flattenImage(image_crop_int)
image_gaus_vec = ImageProcProxy.flattenImage(image_gaus_int)
input_images[0] = ImageProcProxy.concatImages(image_crop_vec,
image_gaus_vec)
return file_list.shape[0], input_images.shape[1], file_list.shape[0]
def train(self, memory_size, input_size, num_samples, filename,stereo_mode):
self.neural_network.memory_size = memory_size
self.neural_network.input_size = input_size
self.neural_network.AllocateNetworkMemories()
file_list = np.genfromtxt(filename, delimiter=',', names=True, dtype=np.dtype([('image',object),
('label', int),
('x', float),
('y', float),
('z',float),
('rx', float),
('ry', float),
('rz',float),
('timestamp', object)]))
for sample in xrange(file_list.shape[0]):
if not sample%100:
print "iterate:",sample
if not stereo_mode :
orig_image_file = ImageProcProxy.readImageColor(file_list['image'][sample])
image_file = cv2.resize(orig_image_file, (params['input']['width'],params['input']['height']))
image_crop = image_file
else:
image_file = ImageProcProxy.readImageColor(file_list['image'][sample])
image_crop = ImageProcProxy.cropImage(image_file, 0, 0, params['input']['width'], params['input']['height'])
image_gaus = ImageProcProxy.applyGaussian(image_crop,
params['filter']['gaussian_radius'],
params['filter']['gaussian_sigma'])
image_crop_int = ImageProcProxy.convertBGR2INT(image_crop)
image_gaus_int = ImageProcProxy.convertBGR2INT(image_gaus)
image_crop_vec = ImageProcProxy.flattenImage(image_crop_int)
image_gaus_vec = ImageProcProxy.flattenImage(image_gaus_int)
input_image = ImageProcProxy.concatImages(image_crop_vec, image_gaus_vec)
input_class = file_list['label'][sample]
self.neural_network.Train( input_image, input_class, sample)
def ProcessImage(image_crop, pixel_range=0):
x = uniform(-pixel_range, pixel_range)
y = uniform(-pixel_range, pixel_range)
image_trans = ImageProcProxy.translateImage(image_crop, x, y)
#ImageProcProxy.showImageBGR(image_trans)
image_gaus = ImageProcProxy.applyGaussian(
image_trans, params['filter']['gaussian_radius'],
params['filter']['gaussian_sigma'])
image_crop_int = ImageProcProxy.convertBGR2INT(image_crop)
image_gaus_int = ImageProcProxy.convertBGR2INT(image_gaus)
image_crop_vec = ImageProcProxy.flattenImage(image_crop_int)
image_gaus_vec = ImageProcProxy.flattenImage(image_gaus_int)
input_image = ImageProcProxy.concatImages(image_crop_vec, image_gaus_vec)
return input_image
def ProcessImage(image_crop, pixel_range=0):
x = uniform(-pixel_range, pixel_range)
y = uniform(-pixel_range, pixel_range)
image_trans = ImageProcProxy.translateImage(image_crop, x, y)
#ImageProcProxy.showImageBGR(image_trans)
image_gaus = ImageProcProxy.applyGaussian(image_trans,
params['filter']['gaussian_radius'],
params['filter']['gaussian_sigma'])
image_crop_int = ImageProcProxy.convertBGR2INT(image_crop)
image_gaus_int = ImageProcProxy.convertBGR2INT(image_gaus)
image_crop_vec = ImageProcProxy.flattenImage(image_crop_int)
image_gaus_vec = ImageProcProxy.flattenImage(image_gaus_int)
input_image = ImageProcProxy.concatImages(image_crop_vec, image_gaus_vec)
return input_image
def LoadDataset(filename, imagepath):
file_list = np.genfromtxt(filename, delimiter=',', names=True, dtype=np.dtype([('timestamp', object), ('x', float), ('y', float), ('label', int)]))
input_images = ImageProcProxy.emptyDataset(file_list.shape[0],
params['input']['width'] * 2,
params['input']['height'])
for sample in xrange(file_list.shape[0]):
image_file = ImageProcProxy.readImageColor(imagepath + file_list['timestamp'][sample] + '.bb08.l.png')
image_crop = ImageProcProxy.cropImage(image_file, 0, 0,
params['input']['width'],
params['input']['height'])
image_gaus = ImageProcProxy.applyGaussian(image_crop,
params['filter']['gaussian_radius'],
params['filter']['gaussian_sigma'])
image_crop_int = ImageProcProxy.convertBGR2INT(image_crop)
image_gaus_int = ImageProcProxy.convertBGR2INT(image_gaus)
image_crop_vec = ImageProcProxy.flattenImage(image_crop_int)
image_gaus_vec = ImageProcProxy.flattenImage(image_gaus_int)
input_images[sample]= ImageProcProxy.concatImages(image_crop_vec, image_gaus_vec)
return input_images, file_list['label'], file_list['x'], file_list['y']