Unsupervised Continual Semantic Adaptation through Neural Rendering

TL;DR

This paper introduces an unsupervised continual semantic adaptation method using Semantic-NeRF, which enables effective multi-scene adaptation for semantic segmentation without ground-truth labels, maintaining performance across scenes.

Contribution

It proposes a novel Semantic-NeRF framework that fuses segmentation predictions with view-consistent rendering for unsupervised continual adaptation.

Findings

Outperforms voxel-based baseline on ScanNet

Outperforms state-of-the-art unsupervised domain adaptation methods

Effectively maintains performance across multiple scenes

Abstract

An increasing amount of applications rely on data-driven models that are deployed for perception tasks across a sequence of scenes. Due to the mismatch between training and deployment data, adapting the model on the new scenes is often crucial to obtain good performance. In this work, we study continual multi-scene adaptation for the task of semantic segmentation, assuming that no ground-truth labels are available during deployment and that performance on the previous scenes should be maintained. We propose training a Semantic-NeRF network for each scene by fusing the predictions of a segmentation model and then using the view-consistent rendered semantic labels as pseudo-labels to adapt the model. Through joint training with the segmentation model, the Semantic-NeRF model effectively enables 2D-3D knowledge transfer. Furthermore, due to its compact size, it can be stored in a long-term memory and subsequently used to render data from arbitrary viewpoints to reduce forgetting. We evaluate our approach on ScanNet, where we outperform both a voxel-based baseline and a state-of-the-art unsupervised domain adaptation method.