Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Vectorize Equalization #201

Merged
merged 8 commits into from
Apr 5, 2022
Merged

Vectorize Equalization #201

Changes from 4 commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
cd68a4f
Port Equalization to BaseImageAugmentationLayer (non-vectorized)
quantumalaviya Mar 23, 2022
22375f6
Vectorizing Equalization
quantumalaviya Mar 24, 2022
020f339
Vectorize Equalization
quantumalaviya Mar 29, 2022
e5fab4b
Vectorize Equalization
quantumalaviya Mar 29, 2022
9885638
added changes
quantumalaviya Apr 5, 2022
d49c8ce
Update equalization.py
quantumalaviya Apr 5, 2022
b7e6654
introduce equalization correctness tests
LukeWood Apr 5, 2022
279c77c
Reformat equalization
LukeWood Apr 5, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 19 additions & 12 deletions keras_cv/layers/preprocessing/equalization.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,9 +60,15 @@ def equalize_channel(self, image, channel_index):
histogram = tf.histogram_fixed_width(image, [0, 255], nbins=self.bins)

# For the purposes of computing the step, filter out the nonzeros.
nonzero = tf.where(tf.not_equal(histogram, 0))
nonzero_histogram = tf.reshape(tf.gather(histogram, nonzero), [-1])
step = (tf.reduce_sum(nonzero_histogram) - nonzero_histogram[-1]) // (
# Zeroes are replaced by 1410065408 while calculating min to keep shape
quantumalaviya marked this conversation as resolved.
Show resolved Hide resolved
# constant across input sizes for compatibility with vectorized_map
histogram_without_zeroes = tf.where(
tf.equal(histogram, 0),
tf.multiply(tf.ones((), tf.int32), 1410065408),
quantumalaviya marked this conversation as resolved.
Show resolved Hide resolved
quantumalaviya marked this conversation as resolved.
Show resolved Hide resolved
histogram,
)

step = (tf.reduce_sum(histogram) - tf.reduce_min(histogram_without_zeroes)) // (
quantumalaviya marked this conversation as resolved.
Show resolved Hide resolved
self.bins - 1
)

Expand All @@ -81,22 +87,23 @@ def build_mapping(histogram, step):
result = tf.cond(
tf.equal(step, 0),
lambda: image,
lambda: tf.cast(
tf.gather(build_mapping(histogram, step), tf.cast(image, tf.int32)),
self.compute_dtype,
),
lambda: tf.gather(build_mapping(histogram, step), image),
)

return result

def augment_image(self, image, transformation=None):
dtype = image.dtype
quantumalaviya marked this conversation as resolved.
Show resolved Hide resolved
image = tf.cast(image, tf.int32)
image = preprocessing.transform_value_range(image, self.value_range, (0, 255))
r = self.equalize_channel(image, 0)
g = self.equalize_channel(image, 1)
b = self.equalize_channel(image, 2)
image = tf.stack([r, g, b], axis=-1)
image = tf.vectorized_map(
lambda channel: self.equalize_channel(image, channel),
tf.range(tf.shape(image)[-1]),
)

image = tf.transpose(image, [1, 2, 0])
image = preprocessing.transform_value_range(image, (0, 255), self.value_range)
return image
return tf.cast(image, dtype)

def get_config(self):
config = super().get_config()
Expand Down