3D Scene Reconstruction Using Instant-NGP

Abstract

Instant Neural Graphics Primitives (instant- ngp) technology enables the rapid reconstruction of 3D scenes from 2D images. This paper provides a prac- tical guide to the installation, configuration, and exe- cution of the instant-NGP tool, focusing on processing video inputs using the colmap2nerf.py script and the main instant-ngp.exe application. The primary contribution of this work is an experimental analysis of the impact of key data preparation parameters – video frame rate (-video_fps) and scene bounding box size (-aabb_scale) – on processing speed and the final vi- sual quality of the 3D reconstruction. The results offer recommendations for the optimal setting of these param- eters depending on the nature of the input video and the desired output.