def test_smart_resize_errors(self):
with self.assertRaisesRegex(ValueError, 'a tuple of 2 integers'):
preprocessing_image.smart_resize(np.random.random((20, 20, 2)),
size=(10, 5, 3))
with self.assertRaisesRegex(ValueError, 'incorrect rank'):
preprocessing_image.smart_resize(np.random.random((20, 40)),
size=(10, 5))
def test_smart_resize(self): test_input = np.random.random((20, 40, 3)) output = preprocessing_image.smart_resize(test_input, size=(50, 50)) self.assertIsInstance(output, np.ndarray) self.assertListEqual(list(output.shape), [50, 50, 3]) output = preprocessing_image.smart_resize(test_input, size=(10, 10)) self.assertListEqual(list(output.shape), [10, 10, 3]) output = preprocessing_image.smart_resize(test_input, size=(100, 50)) self.assertListEqual(list(output.shape), [100, 50, 3]) output = preprocessing_image.smart_resize(test_input, size=(5, 15)) self.assertListEqual(list(output.shape), [5, 15, 3])
def test_smart_resize_tf_dataset(self, size):
test_input_np = np.random.random((2, 20, 40, 3))
test_ds = Dataset.from_tensor_slices(test_input_np)
resize = lambda img: preprocessing_image.smart_resize(img, size=size)
test_ds = test_ds.map(resize)
for sample in test_ds.as_numpy_iterator():
self.assertIsInstance(sample, np.ndarray)
self.assertListEqual(list(sample.shape), [size[0], size[1], 3])
def load_image(path, image_size, num_channels, interpolation,
smart_resize=False):
"""Load an image from a path and resize it."""
img = io_ops.read_file(path)
img = image_ops.decode_image(
img, channels=num_channels, expand_animations=False)
if smart_resize:
img = keras_image_ops.smart_resize(img, image_size,
interpolation=interpolation)
else:
img = image_ops.resize_images_v2(img, image_size, method=interpolation)
img.set_shape((image_size[0], image_size[1], num_channels))
return img
def test_smart_resize_batch(self):
img = np.random.random((2, 20, 40, 3))
out = preprocessing_image.smart_resize(img, size=(20, 20))
self.assertListEqual(list(out.shape), [2, 20, 20, 3])
self.assertAllClose(out, img[:, :, 10:-10, :])
save_plots = True
target_size = 100
# download dataset from http://vis-www.cs.umass.edu/lfw/lfw.tgz
images = glob.glob(os.path.join('data', 'lfw', '**', '*.jpg'))
hi_res_dataset = []
lo_res_dataset = []
ratio = 0.40
low_res_dim = (int(target_size * ratio), int(target_size * ratio))
# for i in tqdm(images[:1000]):
for i in tqdm(images):
hr_img = image.load_img(i, target_size=(target_size, target_size, 3))
hr_img = image.img_to_array(hr_img)
lr_img = smart_resize(hr_img, size=low_res_dim)
lr_img = smart_resize(lr_img, size=(target_size, target_size))
hr_img = hr_img / 255.
lr_img = lr_img / 255.
hi_res_dataset.append(hr_img)
lo_res_dataset.append(lr_img)
hi_res_dataset = np.array(hi_res_dataset)
lo_res_dataset = np.array(lo_res_dataset)
# Hyperparameters
learning_rate = 0.001
n_epochs = 50
batch_size = 256
validation_split = 0.1
activation = 'relu'
