InverseForm: A Loss Function for Structured Boundary-Aware Segmentation

TL;DR

InverseForm introduces a boundary-aware loss function for semantic segmentation that improves boundary accuracy and overall performance without increasing model complexity, validated across multiple benchmarks.

Contribution

The paper proposes a novel inverse-transformation boundary-aware loss that enhances segmentation accuracy by explicitly modeling boundary transformations.

Findings

Consistent performance improvements on Cityscapes, NYU-Depth-v2, and PASCAL datasets.

Sets new state-of-the-art results on two benchmarks.

Effective integration with various backbone networks in different settings.

Abstract

We present a novel boundary-aware loss term for semantic segmentation using an inverse-transformation network, which efficiently learns the degree of parametric transformations between estimated and target boundaries. This plug-in loss term complements the cross-entropy loss in capturing boundary transformations and allows consistent and significant performance improvement on segmentation backbone models without increasing their size and computational complexity. We analyze the quantitative and qualitative effects of our loss function on three indoor and outdoor segmentation benchmarks, including Cityscapes, NYU-Depth-v2, and PASCAL, integrating it into the training phase of several backbone networks in both single-task and multi-task settings. Our extensive experiments show that the proposed method consistently outperforms baselines, and even sets the new state-of-the-art on two datasets.