RGPNet: A Real-Time General Purpose Semantic Segmentation

TL;DR

RGPNet is a lightweight, real-time semantic segmentation model that balances speed and accuracy, utilizing an innovative adaptor and optimized training techniques to outperform heavier models in complex environments.

Contribution

The paper introduces RGPNet, a novel real-time segmentation architecture with an adaptor for improved feature refinement and gradient flow, achieving a superior speed-accuracy trade-off.

Findings

RGPNet achieves real-time segmentation with accuracy comparable to heavy models.

Optimized label-relaxation and progressive resizing reduce training time by up to 60%.

RGPNet performs well across multiple datasets.

Abstract

We propose a real-time general purpose semantic segmentation architecture, RGPNet, which achieves significant performance gain in complex environments. RGPNet consists of a light-weight asymmetric encoder-decoder and an adaptor. The adaptor helps preserve and refine the abstract concepts from multiple levels of distributed representations between the encoder and decoder. It also facilitates the gradient flow from deeper layers to shallower layers. Our experiments demonstrate that RGPNet can generate segmentation results in real-time with comparable accuracy to the state-of-the-art non-real-time heavy models. Moreover, towards green AI, we show that using an optimized label-relaxation technique with progressive resizing can reduce the training time by up to 60% while preserving the performance. We conclude that RGPNet obtains a better speed-accuracy trade-off across multiple datasets.