Semantic Image Segmentation: Tools for New ML models

Source: Deep Learning on Medium

Semantic Image Segmentation: Tools for New ML models

Semantic Image Segmentation is Image Annotation where every pixel of the image gets classified to a category.

Think of building a classification model for each pixel in the image with the choice for the label. Not an easy task

Annotating with the pixel-level precision can be a challenging tasks, that is why only a handful of image annotation and data labeling companies include it in their data offering. While object detection, bounding boxes for computer vision is predominant annotation techniques for Machine Learning model, semantic segmentation is a growing practice as it allows precise image analysis. One of the popular use cases were a Kaggle Carvana Masking challenge. As a classic challenge, the idea behind was to mask out backgrounds, noisy parts of the photos that get uploaded in the car-sale website by users? Such photography enhancement strategies is definitely a game changer for photography and web industry. So how to make image segmentation be pixel level and ensure that the foundation of it — polygonal annotation — would translate the full masks with no pixel lost?

Case study Semantic Segmentation for a photography company

The interesting case for this particular segmentation of the image, is defining the layers. While in most of the aerial images, the set classes don’t overlap, i.e. buildings, vegetation, road, for the case of the photography segmentation, we define classes using indexing.

  • Background — or 0 — index class.
  • Plates, packaging, bowls as a class — or — 1 — level class.
  • Food as a class — or 2- level class.

dWe tested using bounding box annotation for the entire background depicting all the pixels, following the polygonal annotation of the classes with the idea of indexing of overlapping points of objects on the image. By using indexing vs calculation of the x/y of the objects any street level, 90% angle images or full aerial view can be marked for annotation allowing to in a necessary case subtract pixel of a higher level from the lower class pixel.