def on_epoch_end(self, data: Data):
device = next(self.model.parameters()).device
for _ in range(self.trials):
img_path = self.dataset.one_shot_trial(self.N)
input_img = (
np.array([
np.expand_dims(cv2.imread(i, cv2.IMREAD_GRAYSCALE),
-1).reshape((1, 105, 105)) / 255.
for i in img_path[0]
],
dtype=np.float32),
np.array([
np.expand_dims(cv2.imread(i, cv2.IMREAD_GRAYSCALE),
-1).reshape((1, 105, 105)) / 255.
for i in img_path[1]
],
dtype=np.float32))
input_img = (to_tensor(input_img[0], "torch").to(device),
to_tensor(input_img[1], "torch").to(device))
model = self.model.module if torch.cuda.device_count(
) > 1 else self.model
prediction_score = feed_forward(
model, input_img, training=False).cpu().detach().numpy()
if np.argmax(
prediction_score) == 0 and prediction_score.std() > 0.01:
self.correct += 1
self.total += 1
data.write_with_log(self.outputs[0], self.correct / self.total)
def setUpClass(self):
self.pytorch_array = to_tensor(
np.arange(0.0, 8.0, 1.0, dtype=np.float32).reshape(
(1, 1, 2, 2, 2)), 'torch')
self.tensorflow_array = to_tensor(
np.arange(0.0, 8.0, 1.0, dtype=np.float32).reshape(
(1, 2, 2, 2, 1)), 'tf')
def test_normalize_torch_float(self):
op = Normalize(inputs="image",
outputs="image",
mean=0.482,
std=0.289,
max_pixel_value=27.0)
data = op.forward(data=to_tensor(self.numpy_array, "torch"), state={})
np.testing.assert_array_almost_equal(data.numpy(),
self.expected_result, 2)
def test_normalize_torch_multi(self):
op = Normalize(inputs="image",
outputs="image",
mean=(0.44, 0.48, 0.52),
std=(0.287, 0.287, 0.287),
max_pixel_value=27)
data = op.forward(data=to_tensor(self.numpy_array, "torch"), state={})
np.testing.assert_array_almost_equal(data.numpy(),
self.expected_result_multi, 2)
def test_normalize_torch_value_int(self):
np.testing.assert_array_almost_equal(
normalize(to_tensor(self.numpy_array_int, 'torch'), 0.5,
0.31382295, 11.0).numpy(), self.expected_result)
