GeoRoute: Geometry-Aware Hybrid Inference for Traffic Future-Frame Prediction

2026-08-10

AI Summary

Long-term traffic scene prediction often suffers from temporal ghosting, geometric drift, and motion inconsistency, leading to unstable static structures and temporally incoherent outputs. To address this, this work proposes a training-free inference framework that enhances the outputs of pretrained video diffusion models through geometry-aware refinement during inference. The approach uniquely integrates multi-frame depth-layered rendering with vision-language model–guided view-conditioned routing, leveraging a frozen visual language model to achieve geometric stability and cross-view generalization without fine-tuning. Evaluated on the AI City Challenge Track 5, the method achieves state-of-the-art performance, significantly improving fidelity of static structures and consistency of low-level details.

Abstract

Long-horizon future-frame prediction is important for autonomous driving, traffic surveillance, and intelligent transportation systems, yet remains challenging due to temporal ghosting, geometry drift, and inconsistent object motion. Recent latent video diffusion models have achieved impressive visual quality, but directly applying them to structured traffic scenes often leads to unstable geometry and degraded temporal coherence over extended horizons. We present a training-free inference framework that stabilizes reliable static structure in pretrained video predictions through multi-frame temporal context and view-conditioned routing. For front-camera videos, our method refines generated futures with a multi-frame depth-layered renderer that projects static geometry from observed history frames while preserving dynamic regions from the generative base model. For heterogeneous traffic views, a frozen vision-language model infers a coarse camera group from the observed clip and selects a specialized motion-based predictor. The framework requires neither retraining nor fine-tuning of the underlying video model and can be applied directly to pretrained generators. We validate the proposed framework on the AI City Challenge Track 5 benchmark, where our final system achieves competitive performance among the top-ranked teams. These results demonstrate that geometry-aware inference-time refinement and view-conditioned hybrid inference can improve static-geometry stability and low-level structural fidelity without changing the original model architecture.

Problem

• future-frame prediction
• geometry drift
• temporal coherence
• traffic scene
• video generation

Innovation

• geometry-aware inference
• training-free refinement
• multi-frame depth rendering
• view-conditioned routing
• hybrid prediction