To Adapt or Not to Adapt? Real-Time Adaptation for Semantic Segmentation

This study presents HAMLET, a novel framework aiming to overcome the challenges of real-time, online domain adaptation in semantic segmentation. It addresses the problem of high computational costs and inefficiency in handling unforeseeable domain changes, such as sudden weather events. The goal is to enhance the performance, accuracy, and speed of semantic segmentation, while simultaneously adapting at over 29 frames per second (FPS) on a consumer-grade GPU.

 

Publication date: 27 July 2023
Project Page: https://marcbotet.github.io/hamlet-web/
Paper: https://arxiv.org/pdf/2307.15063.pdf