Semantic Segmentation TCGE Framework

Segmentation TGCE is a tool for semantic segmentation of images in the context of multiple annotators.

This tools performs a segmentation task by learning from different sources (annotators) and combining them to obtain a better segmentation. The training algorithm does not require performance metrics from the annotators as an input, but it learns the annotators’ performance from the data, which behaves as a latent variable and regularizes the learning process for “learning” more from the most reliable annotators.

Once learnt, the model can be used to predict the segmentation of new images with a better performance than any of the individual annotators.

Learn more:

• Introduction
  • Why image segmentation?
  • What does multiple annotators mean?
• Loss for multiple annotators segmentation
  • Truncated segmentation generalized cross entropy loss
• Convolutional neural networks
• Experimental setup
  • Oxford III Pet
  • Crowd Seg Histopatological images
• Contribution
  • Core
  • Docs
  • Notebooks

Indices and tables

• Index

• Search Page