CVLGOct 7, 2021

Adaptive Early-Learning Correction for Segmentation from Noisy Annotations

arXiv:2110.03740v2131 citationsHas Code
Originality Highly original
AI Analysis

This addresses the challenge of noisy annotations in segmentation tasks, particularly for medical imaging and weakly-supervised semantic segmentation, with incremental improvements over existing methods.

The paper tackles the problem of training deep segmentation networks on data with noisy annotations by discovering that networks initially fit clean labels before memorizing false ones, and proposes an adaptive early-learning correction method that outperforms standard approaches on medical imaging and achieves state-of-the-art results on PASCAL VOC 2012.

Deep learning in the presence of noisy annotations has been studied extensively in classification, but much less in segmentation tasks. In this work, we study the learning dynamics of deep segmentation networks trained on inaccurately-annotated data. We discover a phenomenon that has been previously reported in the context of classification: the networks tend to first fit the clean pixel-level labels during an "early-learning" phase, before eventually memorizing the false annotations. However, in contrast to classification, memorization in segmentation does not arise simultaneously for all semantic categories. Inspired by these findings, we propose a new method for segmentation from noisy annotations with two key elements. First, we detect the beginning of the memorization phase separately for each category during training. This allows us to adaptively correct the noisy annotations in order to exploit early learning. Second, we incorporate a regularization term that enforces consistency across scales to boost robustness against annotation noise. Our method outperforms standard approaches on a medical-imaging segmentation task where noises are synthesized to mimic human annotation errors. It also provides robustness to realistic noisy annotations present in weakly-supervised semantic segmentation, achieving state-of-the-art results on PASCAL VOC 2012. Code is available at https://github.com/Kangningthu/ADELE

Code Implementations2 repos
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes