PointForward: Feedforward Driving Reconstruction through Point-Aligned Representations

2026-05-12

AI Summary

This work addresses the challenges of multi-view inconsistency, layered artifacts, and the lack of explicit cross-view correspondences for dynamic instances in feed-forward driving scene reconstruction. To this end, we propose PointForward, a novel framework that initializes sparse 3D query points in world coordinates and fuses spatio-temporal multi-view image features to enforce explicit cross-view consistency within a single forward pass. By incorporating a scene graph structure and leveraging 3D bounding boxes, our method organizes dynamic instances and propagates their motion to ensure temporally coherent reconstruction. Replacing the conventional pixel-alignment paradigm with a pioneering point-alignment representation, PointForward substantially mitigates multi-view inconsistencies and dynamic modeling distortions, achieving state-of-the-art performance on large-scale autonomous driving reconstruction benchmarks.

Abstract

High-fidelity reconstruction of driving scenes is crucial for autonomous driving. While recent feedforward 3D Gaussian Splatting (3DGS) methods enable fast reconstruction, their per-pixel Gaussian prediction paradigm often suffers from multi-view inconsistency and layering artifacts. Moreover, existing methods often model dynamic instances via dense flow prediction, which lacks explicit cross-view correspondence and instance-level consistency. In this paper, we propose PointForward, a feedforward driving reconstruction framework through point-aligned representations. Unlike pixel-aligned methods, we initialize sparse 3D queries in world space and aggregate multi-view image information via spatial-temporal fusion onto these queries, enforcing explicit cross-view consistency in a single feedforward pass. To handle scene dynamics, we introduce scene graphs that explicitly organize moving instances during reconstruction. By leveraging 3D bounding boxes, our method enables instance-level motion propagation and temporally consistent dynamic representations. Extensive experiments demonstrate that PointForward achieves state-of-the-art performance on large-scale driving benchmarks. The code will be available upon the publication of the paper.

Problem

• multi-view inconsistency
• layering artifacts
• dynamic instance modeling
• cross-view correspondence
• instance-level consistency

Innovation

• point-aligned representation
• feedforward reconstruction
• 3D Gaussian Splatting
• scene graph
• dynamic instance modeling