MetaBox+: A new Region Based Active Learning Method for Semantic Segmentation using Priority Maps

TL;DR

MetaBox+ introduces a region-based active learning approach for semantic segmentation that estimates segment quality and annotation costs, significantly reducing labeling effort while maintaining high segmentation accuracy.

Contribution

It proposes a meta regression model for segment IoU estimation and a practical annotation cost estimation method, improving active learning efficiency in semantic segmentation.

Findings

Achieves 95% of full dataset mIoU with only 10.47% and 32.01% annotation effort.

Outperforms entropy-based uncertainty methods in reducing annotation costs.

Demonstrates robustness and effectiveness across different networks on Cityscapes.

Abstract

We present a novel region based active learning method for semantic image segmentation, called MetaBox+. For acquisition, we train a meta regression model to estimate the segment-wise Intersection over Union (IoU) of each predicted segment of unlabeled images. This can be understood as an estimation of segment-wise prediction quality. Queried regions are supposed to minimize to competing targets, i.e., low predicted IoU values / segmentation quality and low estimated annotation costs. For estimating the latter we propose a simple but practical method for annotation cost estimation. We compare our method to entropy based methods, where we consider the entropy as uncertainty of the prediction. The comparison and analysis of the results provide insights into annotation costs as well as robustness and variance of the methods. Numerical experiments conducted with two different networks on the Cityscapes dataset clearly demonstrate a reduction of annotation effort compared to random acquisition. Noteworthily, we achieve 95%of the mean Intersection over Union (mIoU), using MetaBox+ compared to when training with the full dataset, with only 10.47% / 32.01% annotation effort for the two networks, respectively.