Reproduce “Foundation4D: Towards a unified perception and reconstruction model for autonomous vehicles” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
NVIDIA, Sérgio Agostinho, Maximilian Apodaca, Emmanuel Attia, Guillem Braso, George Dong, Ilya Fedorov, Michele Fenzi and 19 more
Foundation4D unifies multi-camera video perception and scene reconstruction, predicting depth, camera motion, 3D boxes, semantic occupancy, and renderable Gaussian primitives. Its factorized attention shares across cameras and time, while overlapping windows are assembled with pose-graph optimization and a linear-motion prior. Across 50 clips, this reduces aligned trajectory ATE from 65.55 cm to 31.26 cm, supporting trajectory consistency without resolving long-video drift.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.