cnt += 1
A_file_name = os.path.join(config.base_dir,
config.TRAIN.A_data_dir, cur_dir,
train_file_list[ind] + '.jpg')
B_file_name = os.path.join(config.base_dir,
config.TRAIN.A_data_dir, cur_dir,
true_crop + '.jpg')
# last parameter in file_name, normalized to [0,1]
overlap = float(train_file_list[ind].split("_")[-1]) / 100
if not os.path.isfile(A_file_name) or not os.path.isfile(
B_file_name):
continue
A_image = helper.read_image(A_file_name) # training image A
B_image = helper.read_image(B_file_name) # training image B
feed_dict = {
input_A: A_image,
input_B: B_image,
overlap_input: overlap,
lr: 1e-8
}
# session run
eval = fetcher.fetch(feed_dict, [G_loss])
# evalCX = fetcher.fetch(feed_dict, [CX_content_loss])
if cnt % 100 == 0:
g_loss[ind] = eval[G_loss]
# cx_loss[ind] = evalCX[CX_content_loss]
# ---------------------------------------------------
# train
# ---------------------------------------------------
if config.TRAIN.is_train:
file_list = os.listdir(os.path.join(config.base_dir,config.TRAIN.A_data_dir))
val_file_list = os.listdir(os.path.join(config.base_dir,config.VAL.A_data_dir))
file_list = np.random.permutation(file_list)
assert len(file_list) > 0
train_file_list = file_list[0::config.TRAIN.every_nth_frame]
val_file_list = val_file_list[0::config.VAL.every_nth_frame]
g_loss = np.zeros(len(train_file_list), dtype=float)
fetcher = FetchManager(sess, [G_opt, G_loss])
B_file_name = config.single_image_B_file_name
B_image = helper.read_image(B_file_name) # training image B
## ------------ epoch loop -------------------------
for epoch in range(1, config.TRAIN.num_epochs + 1):
epoch_dir = config.TRAIN.out_dir + "/%04d" % epoch
if os.path.isdir(epoch_dir):
continue
cnt = 0
## ------------ batch loop -------------------------
for ind in np.random.permutation(len(train_file_list)):#
st = time.time()
cnt += 1
A_file_name = config.base_dir + config.TRAIN.A_data_dir + '/' + train_file_list[ind]
if not os.path.isfile(A_file_name) or not os.path.isfile(A_file_name):
def _prewitt(self):
return ([[-1, 0, 1], [-1, 0, 1], [-1, 0, 1]],
[[-1, -1, -1], [0, 0, 0], [1, 1, 1]])
def bounding_box(img, range_w=None, range_h=None):
img = histogram_equalization.adaptive_mean(img)
cv2.imshow('binary', img)
contours, hierarchy = cv2.findContours(img, cv2.RETR_LIST, cv2.CHAIN_APPROX_TC89_KCOS)
boundings = []
for cnt in contours:
x, y, w, h = cv2.boundingRect(cnt)
if (not range_w or (w >= range_w[0] and w <= range_w[1])) and \
(not range_h or (h >= range_h[0] and h <= range_h[1])):
boundings.append((x, y, w, h))
cv2.rectangle(img, (x, y), (x + w, y + h), (255), 2)
return img, boundings
if __name__ == '__main__':
pix = helper.read_image('../resources/fp01.jpg')
pix = helper.convert_gray(pix)
helper.save_image('../resources/gray.png', True)
pix = basic(pix)
# noise_removal = NoiseRemoval()
# pix = noise_removal.opening(pix)
helper.save_image(pix, '../resources/edge.png')