def _net_output_for_word_list(self, word_list, cnn,
min_img_width_height=32, image_size=None, max_pixel=50000, input_layer='word_images',
output_layer='sigmoid', suppress_caffe_output=False):
output = []
for idx, word in enumerate(word_list):
# scale to correct pixel values (0 = background, 1 = text)
word_img = word.get_word_image().astype(np.float32)
word_img -= 255.0
word_img /= -255.0
word_img, resized = self._check_size(word_img,image_size)
word_img = word_img.reshape((1,1,) + word_img.shape).astype(np.float32)
# reshape the PHOCNet
cnn.blobs[input_layer].reshape(*word_img.shape)
cnn.reshape()
# forward the word image through the PHOCNet
cnn.blobs[input_layer].data[...] = word_img
if suppress_caffe_output:
with Suppressor():
output.append(cnn.forward()[output_layer].flatten())
else:
output.append(cnn.forward()[output_layer].flatten())
if ((idx+1)%100 == 0 or (idx+1) == len(word_list)):
self.logger.info(' [ %*d / %d ]', len(str(len(word_list))), idx+1, len(word_list))
return np.vstack(output)
def __solver_step(self, solver, steps):
'''
Runs Caffe solver suppressing Caffe output if necessary
'''
if not self.debug_mode:
with Suppressor():
solver.step(steps)
else:
solver.step(steps)
def _load_pretrained_dense_net(self, phocnet_bin_path, gpu_id, dense_net, debug_mode):
# create the Caffe Dense PHOCNet object
self.logger.info('Creating Dense PHOCNet...')
if debug_mode:
phocnet = caffe.Net(dense_net, phocnet_bin_path,caffe.TEST)
else:
with Suppressor():
phocnet = caffe.Net(dense_net, phocnet_bin_path,caffe.TEST)
return phocnet
def __get_solver(self, solver_proto_path):
'''
Returns a caffe.SGDSolver for the given protofile path,
ignoring Caffe command line chatter if debug mode is not set
to True.
'''
if not self.debug_mode:
# disable Caffe init chatter when not in debug
with Suppressor():
return caffe.SGDSolver(solver_proto_path)
else:
return caffe.SGDSolver(solver_proto_path)
def _load_pretrained_phocnet(self, phocnet_bin_path, gpu_id, debug_mode, deploy_proto_path, phoc_size):
# create a deploy proto file
self.logger.info('Saving PHOCNet deploy proto file to %s...', deploy_proto_path)
mpg = ModelProtoGenerator(initialization='msra', use_cudnn_engine=gpu_id is not None)
proto = mpg.get_phocnet(word_image_lmdb_path=None, phoc_lmdb_path=None, phoc_size=phoc_size, generate_deploy=True)
with open(deploy_proto_path, 'w') as proto_file:
proto_file.write(str(proto))
# create the Caffe PHOCNet object
self.logger.info('Creating PHOCNet...')
if debug_mode:
phocnet = caffe.Net(deploy_proto_path, phocnet_bin_path, caffe.TEST)
else:
with Suppressor():
phocnet = caffe.Net(deploy_proto_path, phocnet_bin_path, caffe.TEST)
return phocnet
def main():
parser = argparse.ArgumentParser()
# required training parameters
parser.add_argument('--net_file', '-f', action='store', type=str,
help='The location of the net file.')
parser.add_argument('--dense', '-d', action='store_true',
help='dense network')
params = vars(parser.parse_args())
caffe.set_mode_cpu()
deploy_file = params["net_file"]
print "Net: " + deploy_file
with Suppressor():
net = caffe.Net(deploy_file, caffe.TEST)
print "Layer-wise parameters: "
pprint([(k, v[0].data.shape, prod(v[0].data.shape) ) for k, v in net.params.items()])
total = sum([prod(v[0].data.shape) for k, v in net.params.items()])
if params["dense"]:
fc = sum([prod(net.params["fc6_d"][0].data.shape),prod(net.params["fc7_d"][0].data.shape),prod(net.params["fc8_d"][0].data.shape)])
else:
fc = sum([prod(net.params["fc6"][0].data.shape),prod(net.params["fc7"][0].data.shape),prod(net.params["fc8"][0].data.shape)])
conv = total-fc
print "Total number of parameters: " + str(total)
print "Convolutional: " + str(conv)
print "FC: " + str(fc)
if params["dense"]:
print "Depth at Tpp: " + str(net.params["fc6_d"][0].data.shape[1]/15)
else:
print "Depth at Tpp: " + str(net.params["fc6"][0].data.shape[1]/15)
def _net_output_for_word_list(self,
word_list,
cnn,
min_img_width_height=26,
input_layer='word_images',
output_layer='sigmoid',
suppress_caffe_output=False):
output = []
for idx, word in enumerate(word_list):
# scale to correct pixel values (0 = background, 1 = text)
word_img = word.get_word_image().astype(np.float32)
word_img -= 255.0
word_img /= -255.0
# check size
if np.amin(word_img.shape[:2]) < min_img_width_height:
scale = float(min_img_width_height + 1) / float(
np.amin(word_img.shape[:2]))
new_shape = (int(scale * word_img.shape[0]),
int(scale * word_img.shape[1]))
word_img = resize(image=word_img, output_shape=new_shape)
word_img = word_img.reshape((
1,
1,
) + word_img.shape).astype(np.float32)
# reshape the PHOCNet
cnn.blobs[input_layer].reshape(*word_img.shape)
cnn.reshape()
# forward the word image through the PHOCNet
cnn.blobs[input_layer].data[...] = word_img
if suppress_caffe_output:
with Suppressor():
output.append(cnn.forward()[output_layer].flatten())
else:
output.append(cnn.forward()[output_layer].flatten())
if ((idx + 1) % 100 == 0 or (idx + 1) == len(word_list)):
self.logger.info(' [ %*d / %d ]', len(str(len(word_list))),
idx + 1, len(word_list))
return np.vstack(output)