def test_op_errors(self):
batch_size = 10
data_height = 9
data_width = 7
data_depth = 3
data_channels = 5
warp_width = 4
warp_height = 8
# Input data shape is not defined over a 2D grid, i.e. its shape is not like
# (batch_size, data_height, data_width, data_channels).
data_shape = (batch_size, data_height, data_width, data_depth,
data_channels)
data = np.zeros(data_shape)
warp_shape = (batch_size, warp_height, warp_width, 2)
warp = np.zeros(warp_shape)
# pylint: disable=bad-continuation
with self.assertRaisesRegexp(
tf.errors.UnimplementedError,
"Only bilinear interpolation is currently supported."):
# pylint: enable=bad-continuation
self.evaluate(resampler_ops.resampler(data, warp))
# Warp tensor must be at least a matrix, with shape [batch_size, 2].
data_shape = (batch_size, data_height, data_width, data_channels)
data = np.zeros(data_shape)
warp_shape = (batch_size, )
warp = np.zeros(warp_shape)
with self.assertRaisesRegexp(tf.errors.InvalidArgumentError,
"warp should be at least a matrix"):
self.evaluate(resampler_ops.resampler(data, warp))
# The batch size of the data and warp tensors must be the same.
data_shape = (batch_size, data_height, data_width, data_channels)
data = np.zeros(data_shape)
warp_shape = (batch_size + 1, warp_height, warp_width, 2)
warp = np.zeros(warp_shape)
with self.assertRaisesRegexp(tf.errors.InvalidArgumentError,
"Batch size of data and warp tensor"):
self.evaluate(resampler_ops.resampler(data, warp))
# The warp tensor must contain 2D coordinates, i.e. its shape last dimension
# must be 2.
data_shape = (batch_size, data_height, data_width, data_channels)
data = np.zeros(data_shape)
warp_shape = (batch_size, warp_height, warp_width, 3)
warp = np.zeros(warp_shape)
# pylint: disable=bad-continuation
with self.assertRaisesRegexp(
tf.errors.UnimplementedError,
"Only bilinear interpolation is supported, warping"):
# pylint: enable=bad-continuation
self.evaluate(resampler_ops.resampler(data, warp))
def test_op_forward_pass(dtype):
np.random.seed(0)
data_width = 7
data_height = 9
data_channels = 5
warp_width = 4
warp_height = 8
batch_size = 10
warp = _make_warp(batch_size, warp_height, warp_width, dtype)
data_shape = (batch_size, data_height, data_width, data_channels)
data = np.random.rand(*data_shape).astype(dtype)
data_ph = tf.constant(data)
warp_ph = tf.constant(warp)
outputs = resampler_ops.resampler(data=data_ph, warp=warp_ph)
assert outputs.shape == (10, warp_height, warp_width, data_channels)
# Generate reference output via bilinear interpolation in numpy
reference_output = np.zeros_like(outputs)
for batch in range(batch_size):
for c in range(data_channels):
reference_output[batch, :, :, c] = _bilinearly_interpolate(
data[batch, :, :, c], warp[batch, :, :, 0], warp[batch, :, :, 1]
)
test_utils.assert_allclose_according_to_type(
outputs, reference_output, half_rtol=5e-3, half_atol=5e-3
)
def _test_op_forward_pass(self, on_gpu, dtype):
np.random.seed(0)
data_width = 7
data_height = 9
data_channels = 5
warp_width = 4
warp_height = 8
batch_size = 10
warp = _make_warp(batch_size, warp_height, warp_width, dtype)
data_shape = (batch_size, data_height, data_width, data_channels)
data = np.random.rand(*data_shape).astype(dtype)
use_gpu = on_gpu and tf.test.is_gpu_available()
with test_utils.device(use_gpu):
data_ph = tf.constant(data)
warp_ph = tf.constant(warp)
outputs = self.evaluate(resampler_ops.resampler(data=data_ph, warp=warp_ph))
self.assertEqual(
outputs.shape, (10, warp_height, warp_width, data_channels)
)
# Generate reference output via bilinear interpolation in numpy
reference_output = np.zeros_like(outputs)
for batch in range(batch_size):
for c in range(data_channels):
reference_output[batch, :, :, c] = _bilinearly_interpolate(
data[batch, :, :, c], warp[batch, :, :, 0], warp[batch, :, :, 1]
)
self.assertAllCloseAccordingToType(
outputs, reference_output, half_rtol=5e-3, half_atol=5e-3
)
def test_op_empty_batch():
np.random.seed(13)
data_width = 5
data_height = 4
data_channels = 3
warp_width = 2
warp_height = 6
batch_size = 0
dtype = np.float32
warp = _make_warp(batch_size, warp_height, warp_width, dtype)
data_shape = (batch_size, data_height, data_width, data_channels)
data = np.zeros(data_shape).astype(dtype)
data_tensor = tf.constant(data)
warp_tensor = tf.constant(warp)
with tf.GradientTape() as tape:
tape.watch(data_tensor)
tape.watch(warp_tensor)
outputs = resampler_ops.resampler(data=data_tensor, warp=warp_tensor)
data_grad, warp_grad = tape.gradient(outputs, (data_tensor, warp_tensor))
assert data_grad.shape == data.shape
assert warp_grad.shape == warp.shape
