Diverse Sampling for Self-Supervised Learning of Semantic Segmentation

TL;DR

This paper introduces a fast, modular self-supervised method for category-level semantic segmentation that leverages localization cues from classification networks, enabling quick training and easy addition of new classes.

Contribution

It presents a nearly hyperparameter-free, efficient approach that automatically labels training points from image-level tags, improving segmentation training speed and flexibility.

Findings

Achieves competitive results on VOC 2012 benchmark.

Training time is less than 3 minutes.

Easily incorporates new classes for inference.

Abstract

We propose an approach for learning category-level semantic segmentation purely from image-level classification tags indicating presence of categories. It exploits localization cues that emerge from training classification-tasked convolutional networks, to drive a "self-supervision" process that automatically labels a sparse, diverse training set of points likely to belong to classes of interest. Our approach has almost no hyperparameters, is modular, and allows for very fast training of segmentation in less than 3 minutes. It obtains competitive results on the VOC 2012 segmentation benchmark. More, significantly the modularity and fast training of our framework allows new classes to efficiently added for inference.