Floxels: Fast Unsupervised Voxel Based Scene Flow Estimation

2025-03-06

AI Summary

Scene flow estimation for autonomous driving faces bottlenecks in unsupervised settings, including low accuracy, poor convergence, and high computational cost. This paper proposes an efficient joint optimization framework for unsupervised scene flow estimation. We introduce a lightweight, voxel-grid-based differentiable parameterization—replacing computationally expensive MLPs—as the first such design. A multi-frame geometric-photometric consistency loss is formulated to strengthen cross-frame constraints and robustness. End-to-end optimization is achieved via differentiable rendering and gradient-based refinement. On Argoverse 2, our method achieves state-of-the-art unsupervised accuracy (second best overall). Inference speed improves by 60–140× over EulerFlow (reducing per-sequence runtime from one day to ten minutes) and by 14× over NSFP, while also delivering significantly superior reconstruction quality. To our knowledge, this is the first unsupervised approach that simultaneously achieves high accuracy, strong robustness, and near-real-time performance.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search
• Machine Learning: Unsupervised & Self-Supervised Learning

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Scene flow estimation is a foundational task for many robotic applications, including robust dynamic object detection, automatic labeling, and sensor synchronization. Two types of approaches to the problem have evolved: 1) Supervised and 2) optimization-based methods. Supervised methods are fast during inference and achieve high-quality results, however, they are limited by the need for large amounts of labeled training data and are susceptible to domain gaps. In contrast, unsupervised test-time optimization methods do not face the problem of domain gaps but usually suffer from substantial runtime, exhibit artifacts, or fail to converge to the right solution. In this work, we mitigate several limitations of existing optimization-based methods. To this end, we 1) introduce a simple voxel grid-based model that improves over the standard MLP-based formulation in multiple dimensions and 2) introduce a new multiframe loss formulation. 3) We combine both contributions in our new method, termed Floxels. On the Argoverse 2 benchmark, Floxels is surpassed only by EulerFlow among unsupervised methods while achieving comparable performance at a fraction of the computational cost. Floxels achieves a massive speedup of more than ~60 - 140x over EulerFlow, reducing the runtime from a day to 10 minutes per sequence. Over the faster but low-quality baseline, NSFP, Floxels achieves a speedup of ~14x.

Problem

• Improves scene flow estimation for robotic applications
• Addresses limitations of unsupervised optimization methods
• Achieves faster runtime with comparable performance

Innovation

• Voxel grid model improves MLP-based formulation
• New multiframe loss formulation enhances accuracy
• Floxels method combines both for speed and quality