Decomposition Sampling for Efficient Region Annotations in Active Learning

TL;DR

This paper introduces Decomposition Sampling (DECOMP), an active learning strategy that efficiently selects diverse, class-specific regions for annotation in dense prediction tasks, significantly improving performance especially on challenging classes.

Contribution

DECOMP is a novel sampling method that decomposes images into class-specific components and guides annotation based on class confidence, reducing computational costs and improving minority-class sampling.

Findings

DECOMP outperforms baseline methods in ROI classification and segmentation tasks.

It effectively samples minority-class regions, enhancing model performance.

DECOMP reduces computational and memory costs compared to existing methods.

Abstract

Active learning improves annotation efficiency by selecting the most informative samples for annotation and model training. While most prior work has focused on selecting informative images for classification tasks, we investigate the more challenging setting of dense prediction, where annotations are more costly and time-intensive, especially in medical imaging. Region-level annotation has been shown to be more efficient than image-level annotation for these tasks. However, existing methods for representative annotation region selection suffer from high computational and memory costs, irrelevant region choices, and heavy reliance on uncertainty sampling. We propose decomposition sampling (DECOMP), a new active learning sampling strategy that addresses these limitations. It enhances annotation diversity by decomposing images into class-specific components using pseudo-labels and sampling regions from each class. Class-wise predictive confidence further guides the sampling process, ensuring that difficult classes receive additional annotations. Across ROI classification, 2-D segmentation, and 3-D segmentation, DECOMP consistently surpasses baseline methods by better sampling minority-class regions and boosting performance on these challenging classes. Code is in https://github.com/JingnaQiu/DECOMP.git.