Location-aware Upsampling for Semantic Segmentation

TL;DR

This paper introduces Location-aware Upsampling (LaU), a trainable module that refines interpolation in semantic segmentation by incorporating location supervision, leading to improved accuracy across benchmarks.

Contribution

The paper proposes a novel trainable upsampling method that integrates location supervision, enhancing segmentation performance over traditional methods.

Findings

Consistent improvement over state-of-the-art methods on benchmark datasets.

LaU can replace standard upsampling modules in encoder-decoder architectures.

End-to-end training of LaU with location-aware losses enhances segmentation accuracy.

Abstract

Many successful learning targets such as minimizing dice loss and cross-entropy loss have enabled unprecedented breakthroughs in segmentation tasks. Beyond these semantic metrics, this paper aims to introduce location supervision into semantic segmentation. Based on this idea, we present a Location-aware Upsampling (LaU) that adaptively refines the interpolating coordinates with trainable offsets. Then, location-aware losses are established by encouraging pixels to move towards well-classified locations. An LaU is offset prediction coupled with interpolation, which is trained end-to-end to generate confidence score at each position from coarse to fine. Guided by location-aware losses, the new module can replace its plain counterpart (\textit{e.g.}, bilinear upsampling) in a plug-and-play manner to further boost the leading encoder-decoder approaches. Extensive experiments validate the consistent improvement over the state-of-the-art methods on benchmark datasets. Our code is available at https://github.com/HolmesShuan/Location-aware-Upsampling-for-Semantic-Segmentation