NAS-based Recursive Stage Partial Network (RSPNet) for Light-Weight Semantic Segmentation

TL;DR

This paper introduces a NAS-based recursive stage partial network (RSPNet) designed for lightweight semantic segmentation, achieving state-of-the-art performance with significantly fewer parameters through a two-stage architecture search and a novel path-attention mechanism.

Contribution

The paper presents a new two-stage NAS framework with a recursive stage partial architecture and path-attention mechanism for efficient lightweight semantic segmentation.

Findings

Achieves state-of-the-art performance on Cityscapes dataset

Uses only 1/4 of parameters compared to SOTA architectures

Completes architecture search in five days on two V100 GPUs

Abstract

Current NAS-based semantic segmentation methods focus on accuracy improvements rather than light-weight design. In this paper, we proposed a two-stage framework to design our NAS-based RSPNet model for light-weight semantic segmentation. The first architecture search determines the inner cell structure, and the second architecture search considers exponentially growing paths to finalize the outer structure of the network. It was shown in the literature that the fusion of high- and low-resolution feature maps produces stronger representations. To find the expected macro structure without manual design, we adopt a new path-attention mechanism to efficiently search for suitable paths to fuse useful information for better segmentation. Our search for repeatable micro-structures from cells leads to a superior network architecture in semantic segmentation. In addition, we propose an RSP (recursive Stage Partial) architecture to search a light-weight design for NAS-based semantic segmentation. The proposed architecture is very efficient, simple, and effective that both the macro- and micro- structure searches can be completed in five days of computation on two V100 GPUs. The light-weight NAS architecture with only 1/4 parameter size of SoTA architectures can achieve SoTA performance on semantic segmentation on the Cityscapes dataset without using any backbones.