Laplacian Pyramid Reconstruction and Refinement for Semantic Segmentation

TL;DR

This paper introduces a Laplacian pyramid-based reconstruction method that refines semantic segmentation boundaries by leveraging multi-resolution features, achieving state-of-the-art results without complex post-processing.

Contribution

It presents a novel multi-resolution reconstruction architecture that refines segmentation boundaries using skip connections and gating, improving accuracy over previous methods.

Findings

Achieves state-of-the-art results on PASCAL VOC and Cityscapes benchmarks.

Demonstrates high sub-pixel localization information in high-dimensional features.

Refines segmentation boundaries effectively without complex inference.

Abstract

CNN architectures have terrific recognition performance but rely on spatial pooling which makes it difficult to adapt them to tasks that require dense, pixel-accurate labeling. This paper makes two contributions: (1) We demonstrate that while the apparent spatial resolution of convolutional feature maps is low, the high-dimensional feature representation contains significant sub-pixel localization information. (2) We describe a multi-resolution reconstruction architecture based on a Laplacian pyramid that uses skip connections from higher resolution feature maps and multiplicative gating to successively refine segment boundaries reconstructed from lower-resolution maps. This approach yields state-of-the-art semantic segmentation results on the PASCAL VOC and Cityscapes segmentation benchmarks without resorting to more complex random-field inference or instance detection driven architectures.