def test_find_local_peaks_rough():
xv, yv = make_grid_vectors(image_height=16,
image_width=16,
output_stride=1)
instances = tf.cast(
[
[[1, 2], [3, 4]],
[[5, 6], [7, 8]],
[[np.nan, np.nan], [11, 12]],
],
tf.float32,
)
cms = make_multi_confmaps(instances, xv=xv, yv=yv, sigma=1.0)
instances2 = tf.cast([[[2, 3], [4, 5]], [[6, 7], [8, 9]]], tf.float32)
cms = tf.stack(
[cms, make_multi_confmaps(instances2, xv=xv, yv=yv, sigma=1.0)],
axis=0)
peak_points, peak_vals, peak_sample_inds, peak_channel_inds = find_local_peaks(
cms, threshold=0.1, refinement=None)
assert peak_points.shape == (9, 2)
assert peak_vals.shape == (9, )
assert peak_sample_inds.shape == (9, )
assert peak_channel_inds.shape == (9, )
assert_array_equal(
peak_points.numpy(),
[
[1, 2],
[3, 4],
[5, 6],
[7, 8],
[11, 12],
[2, 3],
[4, 5],
[6, 7],
[8, 9],
],
)
assert_array_equal(peak_vals, [1, 1, 1, 1, 1, 1, 1, 1, 1])
assert_array_equal(peak_sample_inds, [0, 0, 0, 0, 0, 1, 1, 1, 1])
assert_array_equal(peak_channel_inds, [0, 1, 0, 1, 1, 0, 1, 0, 1])
peak_points, peak_vals, peak_sample_inds, peak_channel_inds = find_local_peaks(
tf.zeros([1, 4, 4, 3], tf.float32), threshold=0.1, refinement=None)
assert peak_points.shape == (0, 2)
assert peak_vals.shape == (0, )
assert peak_sample_inds.shape == (0, )
assert peak_channel_inds.shape == (0, )
def test_centroid_crop_layer():
xv, yv = make_grid_vectors(image_height=12, image_width=12, output_stride=1)
points = tf.cast([[[1.75, 2.75]], [[3.75, 4.75]], [[5.75, 6.75]]], tf.float32)
cms = tf.expand_dims(make_multi_confmaps(points, xv, yv, sigma=1.5), axis=0)
x_in = tf.keras.layers.Input([12, 12, 1])
x_out = tf.keras.layers.Lambda(lambda x: x, name="CentroidConfmapsHead")(x_in)
model = tf.keras.Model(inputs=x_in, outputs=x_out)
layer = CentroidCrop(
keras_model=model,
input_scale=1.0,
crop_size=3,
pad_to_stride=1,
output_stride=None,
refinement="local",
integral_patch_size=5,
peak_threshold=0.2,
return_confmaps=False,
)
out = layer(cms)
assert tuple(out["centroids"].shape) == (1, None, 2)
assert tuple(out["centroid_vals"].shape) == (1, None)
assert tuple(out["crops"].shape) == (1, None, 3, 3, 1)
assert tuple(out["crop_offsets"].shape) == (1, None, 2)
assert tuple(out["centroids"].bounding_shape()) == (1, 3, 2)
assert tuple(out["centroid_vals"].bounding_shape()) == (1, 3)
assert tuple(out["crops"].bounding_shape()) == (1, 3, 3, 3, 1)
assert tuple(out["crop_offsets"].bounding_shape()) == (1, 3, 2)
assert_allclose(out["centroids"][0].numpy(), points.numpy().squeeze(axis=1))
assert_allclose(out["centroid_vals"][0].numpy(), [1, 1, 1], atol=0.1)
def test_find_local_peaks_local():
xv, yv = make_grid_vectors(image_height=32,
image_width=32,
output_stride=1)
instances = (tf.cast(
[
[[1, 2], [3, 4]],
[[5, 6], [7, 8]],
[[np.nan, np.nan], [11, 12]],
],
tf.float32,
) * 2 + 0.25)
cms = make_multi_confmaps(instances, xv=xv, yv=yv, sigma=1.0)
instances2 = tf.cast([[[2, 3], [4, 5]], [[6, 7], [8, 9]]],
tf.float32) * 2 + 0.25
cms = tf.stack(
[cms, make_multi_confmaps(instances2, xv=xv, yv=yv, sigma=1.0)],
axis=0)
peak_points, peak_vals, peak_sample_inds, peak_channel_inds = find_local_peaks(
cms, threshold=0.1, refinement="local")
assert peak_points.shape == (9, 2)
assert peak_vals.shape == (9, )
assert peak_sample_inds.shape == (9, )
assert peak_channel_inds.shape == (9, )
assert_allclose(
peak_points.numpy(),
np.array([
[1, 2],
[3, 4],
[5, 6],
[7, 8],
[11, 12],
[2, 3],
[4, 5],
[6, 7],
[8, 9],
]) * 2 + 0.25,
)
assert_allclose(peak_vals, [1, 1, 1, 1, 1, 1, 1, 1, 1], atol=0.1)
assert_array_equal(peak_sample_inds, [0, 0, 0, 0, 0, 1, 1, 1, 1])
assert_array_equal(peak_channel_inds, [0, 1, 0, 1, 1, 0, 1, 0, 1])
def test_make_multi_confmaps():
xv, yv = make_grid_vectors(image_height=4, image_width=5, output_stride=1)
instances = tf.cast([
[[0.5, 1.], [2., 2.]],
[[1.5, 1.], [2., 3.]],
[[np.nan, np.nan], [-1., 5.]],
], tf.float32)
cms = make_multi_confmaps(instances, xv=xv, yv=yv, sigma=1.)
assert cms.shape == (4, 5, 2)
assert cms.dtype == tf.float32
cm0 = make_confmaps(instances[0], xv=xv, yv=yv, sigma=1.)
cm1 = make_confmaps(instances[1], xv=xv, yv=yv, sigma=1.)
cm2 = make_confmaps(instances[2], xv=xv, yv=yv, sigma=1.)
np.testing.assert_array_equal(
cms, tf.reduce_max(tf.stack([cm0, cm1, cm2], axis=-1), axis=-1))