Fast and Generalizable NeRF Architecture Selection for Satellite Scene Reconstruction

TL;DR

This paper introduces PreSCAN, a fast predictive framework for selecting NeRF architectures for satellite scene reconstruction, significantly reducing search time and enabling deployment on edge devices.

Contribution

We propose PreSCAN, a novel lightweight method that predicts NeRF quality for architecture selection, bypassing costly neural architecture search for satellite imagery.

Findings

PreSCAN achieves <30 seconds architecture selection with <1 dB prediction error.

PreSCAN provides 1000× speedup over traditional NAS methods.

Deployment on edge platforms reduces inference power by 26% and latency by 43%.

Abstract

Neural Radiance Fields (NeRF) have emerged as a powerful approach for photorealistic 3D reconstruction from multi-view images. However, deploying NeRF for satellite imagery remains challenging. Each scene requires individual training, and optimizing architectures via Neural Architecture Search (NAS) demands hours to days of GPU time. While existing approaches focus on architectural improvements, our SHAP analysis reveals that multi-view consistency, rather than model architecture, determines reconstruction quality. Based on this insight, we develop PreSCAN, a predictive framework that estimates NeRF quality prior to training using lightweight geometric and photometric descriptors. PreSCAN selects suitable architectures in < 30 seconds with < 1 dB prediction error, achieving 1000$\times$ speedup over NAS. We further demonstrate PreSCAN's deployment utility on edge platforms (Jetson Orin), where combining its predictions with offline cost profiling reduces inference power by 26% and latency by 43% with minimal quality loss. Experiments on DFC2019 datasets confirm that PreSCAN generalizes across diverse satellite scenes without retraining.