def batch_gen(self, path, batch_size, crop_size):
content_path, mask_path = get_file_paths(path)
while True:
index = random.sample(range(1, len(content_path)), batch_size)
try:
offset_h = random.randint(0, (2448 - crop_size[0]))
offset_w = random.randint(0, (2448 - crop_size[1]))
offset = (offset_h, offset_w)
contents = [
vgg_sub_mean(
random_crop(get_image(content_path[i]), offset,
crop_size)) for i in index
]
masks = [
mask_preprocess(
random_crop(get_image(mask_path[i]), offset,
crop_size)) for i in index
]
contents = np.asarray(contents, dtype=np.float32)
masks = np.asarray(masks, dtype=np.uint8)
except Exception as err:
print("\nError: {}".format(err))
continue
yield contents, masks
def eval(self):
# print('loading model from {}...'.format(self.model.model_path))
# self.model.load(self.model.model_path)
# print('Model is loaded!!!')
self.model.build_model()
path = 'dataset/train/998002_sat.jpg'
img = np.expand_dims(vgg_sub_mean(get_image(path)), axis=0)
reconst_mask = self.model.decode(img)
def eval(self):
# print('loading model from {}...'.format(self.model.model_path))
# self.model.load(self.model.model_path)
# print('Model is loaded!!!')
self.model.build_model()
content_path, _ = get_file_paths(self.test_path)
content_path.sort()
for path in content_path:
fileID = path.split('/')[-1].split('_')[0]
output_name = '{}_mask.png'.format(fileID)
output_name = os.path.join(self.output_path, output_name)
print(output_name)
img = np.expand_dims(vgg_sub_mean(get_image(path)), axis=0)
reconst_mask = self.model.decode(img)
reconst_mask = mask_postprocess(reconst_mask[0])
reconst_mask = image_resize(reconst_mask, size=(612, 612))
skimage.io.imsave(output_name, reconst_mask)
def batch_gen(dir, batch_size):
content_path, mask_path = get_file_paths(dir)
content_path.sort()
mask_path.sort()
while True:
index = random.sample(range(1, len(content_path)), batch_size)
try:
contents = [
vgg_sub_mean(get_image(content_path[i])) for i in index
]
masks = [mask_preprocess(get_image(mask_path[i])) for i in index]
contents = np.asarray(contents, dtype=np.float32)
masks = np.asarray(masks, dtype=np.float32)
except Exception as err:
print("\nError: {}".format(err))
continue
yield contents, masks
def eval(self, output_mode):
# print('loading model from {}...'.format(self.model.model_path))
# self.model.load(self.model.model_path)
# print('Model is loaded!!!')
def merge(group, size=(612, 612), num_per_side=13):
np_mask = np.zeros((2448, 2448, 7))
# img = Image.new('RGB', (2448, 2448))
for idx, chunk in enumerate(group):
offset_x = idx // num_per_side * size[0] // 4
offset_y = idx % num_per_side * size[1] // 4
np_mask[offset_x:offset_x + size[0], offset_y:offset_y +
size[1], :] = np_mask[offset_x:offset_x + size[0],
offset_y:offset_y +
size[1], :] + chunk[:, :, :]
img = mask_postprocess(np_mask)
img = Image.fromarray(img)
return img
self.model.build_model()
content_path, _ = get_file_paths(self.test_path)
content_path.sort()
group = []
for cnt, path in enumerate(content_path):
# assert len(content_path) % 169 == 0
img = np.expand_dims(vgg_sub_mean(get_image(path)), axis=0)
reconst_mask = self.model.decode(img)
if output_mode == 'img':
fileID = path.split('/')[-1].split('_')[0]
output_name = '{}_mask.png'.format(fileID)
output_name = os.path.join(self.output_path, output_name)
print("({}/{}) {}".format(cnt, len(content_path), output_name))
reconst_mask = mask_postprocess(reconst_mask[0])
reconst_mask = image_resize(reconst_mask, size=(612, 612))
skimage.io.imsave(output_name, reconst_mask)
elif output_mode == 'npz':
if cnt % 169 == 0 and cnt != 0:
print('Saving {}'.format(output_name))
img = merge(group)
img.save(output_name)
group = []
fileID = path.split('/')[-1].split('-')[0]
output_name = '{}_mask.png'.format(fileID)
output_name = os.path.join(self.output_path, output_name)
print("({}/{}) {}".format(cnt + 1, len(content_path),
output_name))
scale = 612. / 512.
reconst_mask = scipy.ndimage.interpolation.zoom(
reconst_mask[0], zoom=(scale, scale, 1), mode='reflect')
group.append(reconst_mask)
# Saving the last figure
if output_mode == 'npz':
print('Saving {}'.format(output_name))
img = merge(group)
img.save(output_name)
if __name__ == '__main__':
test_path = args.input_path
output_path = args.output_path
model_path = args.model_path
batch_size = 4
if args.net == 'fcn32':
model = VGG_FCN32(mode='test', model_path=model_path)
elif args.net == 'fcn8':
model = VGG_FCN8(mode='test', model_path=model_path)
content_path = glob.glob("{}/*.jpg".format(test_path))
content_path.sort()
for i, path in enumerate(content_path):
output_name = "{}{:04d}_mask.png".format(output_path, i)
img = vgg_sub_mean(get_image(path))
img = np.expand_dims(img, axis=0)
reconstructed_mask = model.decode(img)
reconstructed_mask = mask_postprocess(reconstructed_mask[0])
skimage.io.imsave(output_name, reconstructed_mask)