TensorFlow Addons Image: Operations

View on TensorFlow.org View source on GitHub Download notebook

Overview

This notebook will demonstrate how to use the some image operations in TensorFlow Addons.

Here is the list of image operations we'll be covering in this example:

Setup

pip install -q -U tensorflow-addons
import tensorflow as tf
import numpy as np
import tensorflow_addons as tfa
import matplotlib.pyplot as plt

Prepare and Inspect Images

Download the images

img_path = tf.keras.utils.get_file('tensorflow.png','https://tensorflow.org/images/tf_logo.png')
Downloading data from https://tensorflow.org/images/tf_logo.png
40960/39781 [==============================] - 0s 2us/step

Inspect the images

TensorFlow Icon

img_raw = tf.io.read_file(img_path)
img = tf.io.decode_image(img_raw)
img = tf.image.convert_image_dtype(img, tf.float32)
img = tf.image.resize(img, [500,500])

plt.title("TensorFlow Logo with shape {}".format(img.shape))
_ = plt.imshow(img)

png

Make a black and white version

bw_img = 1.0 - tf.image.rgb_to_grayscale(img)

plt.title("Mask image with shape {}".format(bw_img.shape))
_ = plt.imshow(bw_img[...,0], cmap='gray')

png

Play with tfa.image

Mean filtering

Mean filtering is a filtering technique, which is often used to remove noise from an image or signal. The idea is to run through the image pixel by pixel and replacing it with the average values of neighboring pixels.

mean = tfa.image.mean_filter2d(img, filter_shape=11)
_ = plt.imshow(mean)

png

Rotate

This operation rotates the given image by the angle (in radians) input by the user.

rotate = tfa.image.rotate(img, tf.constant(np.pi/8))
_ = plt.imshow(rotate)

png

Transform

This operation transforms the given image on the basis of the transform vector given by the user.

transform = tfa.image.transform(img, [1.0, 1.0, -250, 0.0, 1.0, 0.0, 0.0, 0.0])
_ = plt.imshow(transform)

png

Random HSV in YIQ

This operation changes color scale of a given RGB image to YIQ but here delta hue and saturation values are picked randomly from the given range.

delta = 0.5
lower_saturation = 0.1
upper_saturation = 0.9
lower_value = 0.2
upper_value = 0.8
rand_hsvinyiq = tfa.image.random_hsv_in_yiq(img, delta, lower_saturation, upper_saturation, lower_value, upper_value)
_ = plt.imshow(rand_hsvinyiq)
/tmpfs/src/tf_docs_env/lib/python3.6/site-packages/tensorflow_addons/utils/resource_loader.py:103: UserWarning: You are currently using TensorFlow 2.3.0 and trying to load a custom op (custom_ops/image/_distort_image_ops.so).
TensorFlow Addons has compiled its custom ops against TensorFlow 2.2.0, and there are no compatibility guarantees between the two versions. 
This means that you might get segfaults when loading the custom op, or other kind of low-level errors.
 If you do, do not file an issue on Github. This is a known limitation.

It might help you to fallback to pure Python ops with TF_ADDONS_PY_OPS . To do that, see https://github.com/tensorflow/addons#gpucpu-custom-ops 

You can also change the TensorFlow version installed on your system. You would need a TensorFlow version equal to or above 2.2.0 and strictly below 2.3.0.
 Note that nightly versions of TensorFlow, as well as non-pip TensorFlow like `conda install tensorflow` or compiled from source are not supported.

The last solution is to find the TensorFlow Addons version that has custom ops compatible with the TensorFlow installed on your system. To do that, refer to the readme: https://github.com/tensorflow/addons
  UserWarning,

png

Adjust HSV in YIQ

This operation changes color scale of a given RGB image to YIQ but here instead of choosing randomly, delta hue and saturation values are inputs form the user.

delta = 0.5
saturation = 0.3
value = 0.6
adj_hsvinyiq = tfa.image.adjust_hsv_in_yiq(img, delta, saturation, value)
_ = plt.imshow(adj_hsvinyiq)

png

Dense Image Warp

This operation is for non-linear warp of any image specified by the flow field of the offset vector (here used random values for example).

input_img = tf.image.convert_image_dtype(tf.expand_dims(img, 0), tf.dtypes.float32)

flow_shape = [1, input_img.shape[1], input_img.shape[2], 2]
init_flows = np.float32(np.random.normal(size=flow_shape) * 2.0)
dense_img_warp = tfa.image.dense_image_warp(input_img, init_flows)
dense_img_warp = tf.squeeze(dense_img_warp, 0)
_ = plt.imshow(dense_img_warp)

png

Euclidian Distance Transform

This operation updates the pixel value with the euclidian distance from the foreground pixel to the background one.

  • Note : It takes only binary image and results in transformed image. If a different image is given it results in a image with single value
gray = tf.image.convert_image_dtype(bw_img,tf.uint8)
# The op expects a batch of images, so add a batch dimension
gray = tf.expand_dims(gray, 0)
eucid = tfa.image.euclidean_dist_transform(gray)
eucid = tf.squeeze(eucid, (0, -1))
_ = plt.imshow(eucid, cmap='gray')

png