Mining of Single-Class by Active Learning for Semantic Segmentation

TL;DR

MiSiCAL introduces a deep reinforcement learning-based active learning method that efficiently focuses on underrepresented classes in semantic segmentation, reducing retraining needs and outperforming baselines on most classes.

Contribution

The paper presents MiSiCAL, a novel active learning paradigm that leverages fixed data representations and reinforcement learning to target specific classes without repeated model retraining.

Findings

MiSiCAL outperforms random policy on 150 out of 171 classes.

It is especially effective with large batch sizes.

It surpasses the strongest baseline on more classes.

Abstract

Several Active Learning (AL) policies require retraining a target model several times in order to identify the most informative samples and rarely offer the option to focus on the acquisition of samples from underrepresented classes. Here the Mining of Single-Class by Active Learning (MiSiCAL) paradigm is introduced where an AL policy is constructed through deep reinforcement learning and exploits quantity-accuracy correlations to build datasets on which high-performance models can be trained with regards to specific classes. MiSiCAL is especially helpful in the case of very large batch sizes since it does not require repeated model training sessions as is common in other AL methods. This is thanks to its ability to exploit fixed representations of the candidate data points. We find that MiSiCAL is able to outperform a random policy on 150 out of 171 COCO10k classes, while the strongest baseline only outperforms random on 101 classes.