def read_images(fpath):
lines = utils.read_image_list(fpath)
logger.info('loading data: {}'.format(fpath))
X_data, y_data = [], []
for inst_path, truth_path in lines:
inst, truth = [cv2.imread(p, cv2.IMREAD_GRAYSCALE)
for p in (inst_path, truth_path)]
assert inst is not None and truth is not None, (inst_path, truth_path)
pad_h, pad_w = [x / 2 for x in MODEL_INPUT_SHAPE]
padded = cv2.copyMakeBorder(inst, pad_h, pad_h, pad_w, pad_w,
cv2.BORDER_REFLECT)
m7 = cv2.medianBlur(padded, 7)
m15 = cv2.medianBlur(padded, 15)
c7 = 255 - cv2.subtract(m7, padded)
c15 = 255 - cv2.subtract(m15, padded)
# (c, h, w) layout
input = np.array((padded, m7, m15, c7, c15))
truth = truth.reshape((1,) + truth.shape)
# pad input image
X_data.append(input)
y_data.append(truth)
return X_data, y_data
def read_images(fpath):
lines = utils.read_image_list(fpath)
is_train = os.path.basename(fpath) == 'train.list'
logger.info('loading data: {}'.format(fpath))
X_data, y_data = [], []
for inst_path, truth_path in lines:
inst, truth = [cv2.imread(p, cv2.IMREAD_GRAYSCALE)
for p in (inst_path, truth_path)]
assert inst is not None and truth is not None, (inst_path, truth_path)
pad_h, pad_w = [x / 2 for x in MODEL_INPUT_SHAPE]
# pad input image
inst = cv2.copyMakeBorder(inst, pad_h, pad_h, pad_w, pad_w,
cv2.BORDER_REFLECT)
truth_padded = cv2.copyMakeBorder(truth, pad_h, pad_h, pad_w, pad_w,
cv2.BORDER_REFLECT)
X_data.append(inst)
y_data.append(truth)
if is_train:
insts = generate_data(truth_padded)
for i in insts:
# cv2.imshow('', i)
# cv2.waitKey(0)
X_data.append(i)
y_data.append(truth)
return X_data, y_data
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--image_list', required=True)
parser.add_argument('--config', required=True)
parser.add_argument('--dump_prefix', required=True)
parser.add_argument('--output_dir', required=True)
args = parser.parse_args()
conf_mod = imp.load_source('config', args.config)
config = conf_mod.get()
model = config['model']
utils.load_model(model, args.dump_prefix)
X, _ = conf_mod.get_data(args.image_list)
utils.mkdir_p(args.output_dir)
image_list = utils.read_image_list(args.image_list)
logger.info('compiling model ...')
model.compile(loss='mean_squared_error', optimizer=Adam())
for x, (input_path, _) in ProgressBar()(zip(X, image_list)):
y = model.predict(np.array([x], dtype='float32'),
batch_size=1, verbose=False)
img = np.round(y.reshape(y.shape[2:]) * 255.0).astype('uint8')
# FIXME: we assume that basenames of images are distinct
fname = os.path.basename(input_path)
output_path = os.path.join(args.output_dir, fname)
cv2.imwrite(output_path, img)
def GetImage(image_path):
#Get the images from the path of image
list_file = read_image_list(image_path)
list_file.sort(compare)
image_array = getShapeForData(list_file)
return image_array
def get_bgimg_pool():
global _bgimgs
if _bgimgs is None:
list_path = os.path.join(DATA_DIR, 'background.list')
paths = sum(utils.read_image_list(list_path), [])
_bgimgs = [cv2.imread(p, cv2.IMREAD_GRAYSCALE) for p in paths]
_bgimgs = filter(lambda x: x is not None, _bgimgs)
logger.info('{} background images'.format(len(_bgimgs)))
return _bgimgs
def read_images(fpath):
lines = utils.read_image_list(fpath)
logger.info("loading data: {}".format(fpath))
X_data, y_data = [], []
for inst_path, truth_path in lines:
inst, truth = [cv2.imread(p, cv2.IMREAD_GRAYSCALE) for p in (inst_path, truth_path)]
assert inst is not None and truth is not None, (inst_path, truth_path)
pad_h, pad_w = [x / 2 for x in MODEL_INPUT_SHAPE]
inst = cv2.copyMakeBorder(inst, pad_h, pad_h, pad_w, pad_w, cv2.BORDER_REFLECT)
# pad input image
X_data.append(inst)
y_data.append(truth)
return X_data, y_data
def read_images(fpath):
lines = utils.read_image_list(fpath)
logger.info('loading data: {}'.format(fpath))
X_data, y_data = [], []
for inst_path, truth_path in lines:
inst, truth = [cv2.imread(p, cv2.IMREAD_GRAYSCALE)
for p in (inst_path, truth_path)]
assert inst is not None and truth is not None, (inst_path, truth_path)
pad_h, pad_w = [x / 2 for x in MODEL_INPUT_SHAPE]
padded = cv2.copyMakeBorder(inst, pad_h, pad_h, pad_w, pad_w,
cv2.BORDER_REFLECT)
truth = truth.reshape((1,) + truth.shape)
for img in [padded] + [augment(padded) for _ in range(10)]:
input = gen_multi_channel(img)
X_data.append(input)
y_data.append(truth)
return X_data, y_data
def read_images(fpath):
lines = utils.read_image_list(fpath)
is_train = os.path.basename(fpath) == 'train.list'
logger.info('loading data: {}'.format(fpath))
X_data, y_data = [], []
for inst_path, truth_path in lines:
inst, truth = [cv2.imread(p, cv2.IMREAD_GRAYSCALE)
for p in (inst_path, truth_path)]
assert inst is not None and truth is not None, (inst_path, truth_path)
pad_h, pad_w = [x / 2 for x in MODEL_INPUT_SHAPE]
# pad input image
padded = cv2.copyMakeBorder(inst, pad_h, pad_h, pad_w, pad_w,
cv2.BORDER_REFLECT)
truth_padded = cv2.copyMakeBorder(truth, pad_h, pad_h, pad_w, pad_w,
cv2.BORDER_REFLECT)
truth = truth.reshape((1,) + truth.shape)
input = gen_multi_channel(augment(padded))
X_data.append(input)
y_data.append(truth)
if is_train:
insts = generate_data(truth_padded)
for i in insts:
input = gen_multi_channel(augment(i))
# for c in range(5):
# cv2.imshow(str(c), input[c])
# cv2.waitKey(0)
X_data.append(input)
y_data.append(truth)
return X_data, y_data