def test_make_confmaps():
xv, yv = make_grid_vectors(image_height=4, image_width=5, output_stride=1)
points = tf.cast([[0.5, 1.], [3, 3.5], [2., 2.]], tf.float32)
cm = make_confmaps(points, xv, yv, sigma=1.)
assert cm.dtype == tf.float32
assert cm.shape == (4, 5, 3)
np.testing.assert_allclose(cm, [[
[0.535, 0., 0.018],
[0.535, 0., 0.082],
[0.197, 0.001, 0.135],
[0.027, 0.002, 0.082],
[0.001, 0.001, 0.018],
],
[
[0.882, 0., 0.082],
[0.882, 0.006, 0.368],
[0.325, 0.027, 0.607],
[0.044, 0.044, 0.368],
[0.002, 0.027, 0.082],
],
[
[0.535, 0.004, 0.135],
[0.535, 0.044, 0.607],
[0.197, 0.197, 1.],
[0.027, 0.325, 0.607],
[0.001, 0.197, 0.135],
],
[
[0.119, 0.01, 0.082],
[0.119, 0.119, 0.368],
[0.044, 0.535, 0.607],
[0.006, 0.882, 0.368],
[0., 0.535, 0.082],
]],
atol=1e-3)
# Grid aligned peak
points = tf.cast([[2, 3]], tf.float32)
cm = make_confmaps(points, xv, yv, sigma=1.)
assert cm.shape == (4, 5, 1)
assert cm[3, 2] == 1.0
# Output stride
xv, yv = make_grid_vectors(image_height=8, image_width=8, output_stride=2)
points = tf.cast([[2, 4]], tf.float32)
cm = make_confmaps(points, xv, yv, sigma=1.)
assert cm.shape == (4, 4, 1)
assert cm[2, 1] == 1.0
# Missing points
xv, yv = make_grid_vectors(image_height=8, image_width=8, output_stride=2)
points = tf.cast([[2, 4]], tf.float32)
cm = make_confmaps(points, xv, yv, sigma=1.)
points_with_nan = tf.cast([[2, 4], [np.nan, np.nan]], tf.float32)
cm_with_nan = make_confmaps(points_with_nan, xv, yv, sigma=1.)
assert cm_with_nan.shape == (4, 4, 2)
assert cm_with_nan.dtype == tf.float32
np.testing.assert_array_equal(cm_with_nan[:, :, 0], cm[:, :, 0])
assert (cm_with_nan[:, :, 1].numpy() == 0).all()
def test_single_instance_inference():
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)
points = np.stack([points, points + 1], axis=0)
cms = tf.stack(
[
make_confmaps(points[0], xv, yv, sigma=1.0),
make_confmaps(points[1], xv, yv, sigma=1.0),
],
axis=0,
)
x_in = tf.keras.layers.Input([12, 12, 3])
x = tf.keras.layers.Lambda(lambda x: x, name="SingleInstanceConfmapsHead")(x_in)
keras_model = tf.keras.Model(x_in, x)
layer = SingleInstanceInferenceLayer(keras_model=keras_model, refinement="local")
assert layer.output_stride == 1
out = layer(cms)
assert tuple(out["instance_peaks"].shape) == (2, 1, 3, 2)
out["instance_peaks"] = tf.squeeze(out["instance_peaks"], axis=1)
assert tuple(out["instance_peak_vals"].shape) == (2, 1, 3)
out["instance_peak_vals"] = tf.squeeze(out["instance_peak_vals"], axis=1)
assert_array_equal(out["instance_peaks"], points)
assert_allclose(out["instance_peak_vals"], 1.0, atol=0.1)
assert "confmaps" not in out
out = layer({"image": cms})
assert tuple(out["instance_peaks"].shape) == (2, 1, 3, 2)
out["instance_peaks"] = tf.squeeze(out["instance_peaks"], axis=1)
assert_array_equal(out["instance_peaks"], points)
layer = SingleInstanceInferenceLayer(
keras_model=keras_model, refinement="local", return_confmaps=True
)
out = layer(cms)
assert "confmaps" in out
assert_array_equal(out["confmaps"], cms)
model = SingleInstanceInferenceModel(layer)
preds = model.predict(cms)
assert preds["instance_peaks"].shape == (2, 1, 3, 2)
preds["instance_peaks"] = preds["instance_peaks"].squeeze(axis=1)
assert_array_equal(preds["instance_peaks"], points)
assert "instance_peak_vals" in preds
assert "confmaps" in preds
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))
def test_find_global_peaks_rough():
xv, yv = make_grid_vectors(image_height=8, image_width=8, output_stride=1)
points = tf.cast([[1, 2], [3, 4], [5, 6]], tf.float32)
cm = make_confmaps(points, xv, yv, sigma=1.0)
points2 = points + 1
cms = tf.stack([cm, make_confmaps(points2, xv, yv, sigma=1.0)])
peaks, peak_vals = find_global_peaks(cms, threshold=0.1, refinement=None)
assert peaks.shape == (2, 3, 2)
assert peak_vals.shape == (2, 3)
assert_array_equal(peaks[0], points)
assert_array_equal(peak_vals[0], [1, 1, 1])
assert_array_equal(peaks[1], points2)
assert_array_equal(peak_vals[1], [1, 1, 1])
peaks, peak_vals = find_global_peaks_rough(tf.zeros((1, 8, 8, 3),
dtype=tf.float32),
threshold=0.1)
assert peaks.shape == (1, 3, 2)
assert peak_vals.shape == (1, 3)
assert tf.reduce_all(tf.math.is_nan(peaks))
assert_array_equal(peak_vals, [[0, 0, 0]])
def test_find_global_peaks_integral():
xv, yv = make_grid_vectors(image_height=12,
image_width=12,
output_stride=1)
points = tf.cast([[1.5, 2.5], [3.5, 4.5], [5.5, 6.5]], tf.float32)
cm = make_confmaps(points, xv, yv, sigma=1.0)
peaks, peak_vals = find_global_peaks(
tf.expand_dims(cm, axis=0),
threshold=0.1,
refinement="integral",
integral_patch_size=5,
)
assert peaks.shape == (1, 3, 2)
assert peak_vals.shape == (1, 3)
assert_allclose(peaks[0].numpy(), points.numpy(), atol=0.1)
assert_allclose(peak_vals[0].numpy(), [1, 1, 1], atol=0.3)
peaks, peak_vals = find_global_peaks(
tf.zeros((1, 8, 8, 3), dtype=tf.float32),
threshold=0.1,
refinement="integral",
integral_patch_size=5,
)
assert peaks.shape == (1, 3, 2)
assert peak_vals.shape == (1, 3)
assert tf.reduce_all(tf.math.is_nan(peaks))
assert_array_equal(peak_vals, [[0, 0, 0]])
peaks, peak_vals = find_global_peaks(
tf.stack([tf.zeros([12, 12, 3], dtype=tf.float32), cm], axis=0),
threshold=0.1,
refinement="integral",
integral_patch_size=5,
)
assert peaks.shape == (2, 3, 2)
assert tf.reduce_all(tf.math.is_nan(peaks[0]))
assert_allclose(peaks[1].numpy(), points.numpy(), atol=0.1)
peaks, peak_vals = find_global_peaks_integral(
tf.stack([tf.zeros([12, 12, 3], dtype=tf.float32), cm], axis=0),
threshold=0.1,
crop_size=5,
)
assert peaks.shape == (2, 3, 2)
assert tf.reduce_all(tf.math.is_nan(peaks[0]))
assert_allclose(peaks[1].numpy(), points.numpy(), atol=0.1)
def test_find_global_peaks_local():
xv, yv = make_grid_vectors(image_height=12,
image_width=12,
output_stride=1)
points = tf.cast([[1.6, 2.6], [3.6, 4.6], [5.6, 6.6]], tf.float32)
cm = make_confmaps(points, xv, yv, sigma=1.0)
peaks, peak_vals = find_global_peaks(tf.expand_dims(cm, axis=0),
threshold=0.1,
refinement="local")
assert peaks.shape == (1, 3, 2)
assert peak_vals.shape == (1, 3)
assert_allclose(peaks[0].numpy(),
np.array([[1.75, 2.75], [3.75, 4.75], [5.75, 6.75]]))
assert_allclose(peak_vals[0].numpy(), [1, 1, 1], atol=0.3)
def test_instance_peaks_layer():
xv, yv = make_grid_vectors(image_height=12, image_width=12, output_stride=1)
points = tf.cast([[1.5, 2.5], [3.5, 4.5], [5.5, 6.5]], tf.float32)
cms = tf.stack(
[
make_confmaps(points, xv, yv, sigma=1.0),
make_confmaps(points + 1, xv, yv, sigma=1.0),
],
axis=0,
)
x_in = tf.keras.layers.Input([12, 12, 3])
x_out = tf.keras.layers.Lambda(lambda x: x, name="CenteredInstanceConfmapsHead")(
x_in
)
model = tf.keras.Model(inputs=x_in, outputs=x_out)
instance_peaks_layer = FindInstancePeaks(
keras_model=model,
input_scale=1.0,
peak_threshold=0.2,
return_confmaps=False,
refinement="integral",
)
# Raw tensor
out = instance_peaks_layer(cms)
assert tuple(out["instance_peaks"].shape) == (2, None, 3, 2)
assert tuple(out["instance_peak_vals"].shape) == (2, None, 3)
assert tuple(out["instance_peaks"].bounding_shape()) == (2, 1, 3, 2)
assert tuple(out["instance_peak_vals"].bounding_shape()) == (2, 1, 3)
assert_allclose(out["instance_peaks"][0][0].numpy(), points.numpy(), atol=0.1)
assert_allclose(out["instance_peak_vals"][0][0].numpy(), [1, 1, 1], atol=0.3)
assert_allclose(out["instance_peaks"][1][0].numpy(), points.numpy() + 1, atol=0.1)
assert_allclose(out["instance_peak_vals"][1][0].numpy(), [1, 1, 1], atol=0.3)
# Batched example
crops = tf.RaggedTensor.from_tensor(tf.expand_dims(cms, axis=1), lengths=[1, 1])
out = instance_peaks_layer({"crops": crops})
assert tuple(out["instance_peaks"].shape) == (2, None, 3, 2)
assert tuple(out["instance_peak_vals"].shape) == (2, None, 3)
assert tuple(out["instance_peaks"].bounding_shape()) == (2, 1, 3, 2)
assert tuple(out["instance_peak_vals"].bounding_shape()) == (2, 1, 3)
assert_allclose(out["instance_peaks"][0][0].numpy(), points.numpy(), atol=0.1)
assert_allclose(out["instance_peak_vals"][0][0].numpy(), [1, 1, 1], atol=0.3)
assert_allclose(out["instance_peaks"][1][0].numpy(), points.numpy() + 1, atol=0.1)
assert_allclose(out["instance_peak_vals"][1][0].numpy(), [1, 1, 1], atol=0.3)
# Batch size = 1, multi-instance example
crops = tf.RaggedTensor.from_tensor(tf.expand_dims(cms, axis=0), lengths=[2])
out = instance_peaks_layer({"crops": crops})
assert tuple(out["instance_peaks"].shape) == (1, None, 3, 2)
assert tuple(out["instance_peak_vals"].shape) == (1, None, 3)
assert tuple(out["instance_peaks"].bounding_shape()) == (1, 2, 3, 2)
assert tuple(out["instance_peak_vals"].bounding_shape()) == (1, 2, 3)
assert_allclose(out["instance_peaks"][0][0].numpy(), points.numpy(), atol=0.1)
assert_allclose(out["instance_peak_vals"][0][0].numpy(), [1, 1, 1], atol=0.3)
assert_allclose(out["instance_peaks"][0][1].numpy(), points.numpy() + 1, atol=0.1)
assert_allclose(out["instance_peak_vals"][0][1].numpy(), [1, 1, 1], atol=0.3)
# Offset adjustment and pass through centroids
instance_peaks_layer = FindInstancePeaks(
keras_model=model,
input_scale=1.0,
peak_threshold=0.2,
return_confmaps=True,
refinement="integral",
)
# (samples, h, w, c) -> (samples, ?, h, w, c)
crops = tf.RaggedTensor.from_tensor(tf.expand_dims(cms, axis=1), lengths=[1, 1])
# (samples, centroids, 2) -> (samples, ?, 2)
crop_offsets = tf.RaggedTensor.from_tensor(
tf.reshape(tf.cast([1, 2, 3, 4], tf.float32), [2, 1, 2]), lengths=[1, 1]
)
out = instance_peaks_layer(
{
"crops": crops,
"centroids": tf.zeros([]),
"centroid_vals": tf.zeros([]),
"crop_offsets": crop_offsets,
}
)
assert "centroids" in out
assert "centroid_vals" in out
assert_allclose(
out["instance_peaks"][0][0].numpy(),
points.numpy() + np.array([[1, 2]]),
atol=0.1,
)
assert_allclose(
out["instance_peaks"][1][0].numpy(),
points.numpy() + 1 + np.array([[3, 4]]),
atol=0.1,
)
# Input scaling
scale = 0.5
instance_peaks_layer = FindInstancePeaks(
keras_model=model,
input_scale=scale,
peak_threshold=0.2,
return_confmaps=False,
refinement="integral",
)
xv, yv = make_grid_vectors(
image_height=12 / scale, image_width=12 / scale, output_stride=1
)
points = tf.cast([[1.5, 2.5], [3.5, 4.5], [5.5, 6.5]], tf.float32)
cms = tf.stack(
[
make_confmaps(points / scale, xv, yv, sigma=1.0 / scale),
make_confmaps((points + 1) / scale, xv, yv, sigma=1.0 / scale),
],
axis=0,
)
out = instance_peaks_layer(cms)
assert_allclose(
out["instance_peaks"][0][0].numpy(), points.numpy() / scale, atol=0.15
)
assert_allclose(
out["instance_peaks"][1][0].numpy(), (points.numpy() + 1) / scale, atol=0.15
)