Exciting news! Gradient has launched a FREE GPU plan. Read More
Project Details

Image Segmentation with TensorFlow

Using TensorFlow to determine objects and surroundings in images

The Gradient Team


This tutorial will walk you through image segmentation using a modified U-Net on the Oxford-IIIT Pet Dataset (created by Parkhi et al).

Image segmentation involves training a neural network to output a pixel-wise mask of an image. Each pixel is given a label which determines if it belongs to the object in that image, or not.

The Oxford-IIIT Pet Dataset consists of images, their corresponding labels, and pixel-wise masks. These masks are essentially labels for each pixel, which fall into three categories:

  • Class 1 : Pixel belonging to the pet.
  • Class 2 : Pixel bordering the pet.
  • Class 3 : Surrounding pixel.

Image segmentation has many applications, for example in medical imaging, self-driving cars and satellite imaging.