arXiv:2511.19235v1 Announce Type: new 
Abstract: Reconstructing dynamic driving scenes is essential for developing autonomous systems through sensor-realistic simulation. Although recent methods achieve high-fidelity reconstructions, they either rely on costly human annotations for object trajectories or use time-varying representations without explicit object-level decomposition, leading to intertwined static and dynamic elements that hinder scene separation. We present IDSplat, a self-supervised 3D Gaussian Splatting framework that reconstructs dynamic scenes with explicit instance decomposition and learnable motion trajectories, without requiring human annotations. Our key insight is to model dynamic objects as coherent instances undergoing rigid transformations, rather than unstructured time-varying primitives. For instance decomposition, we employ zero-shot, language-grounded video tracking anchored to 3D using lidar, and estimate consistent poses via feature correspondences. We introduce a coordinated-turn smoothing scheme to obtain temporally and physically consistent motion trajectories, mitigating pose misalignments and tracking failures, followed by joint optimization of object poses and Gaussian parameters. Experiments on the Waymo Open Dataset demonstrate that our method achieves competitive reconstruction quality while maintaining instance-level decomposition and generalizes across diverse sequences and view densities without retraining, making it practical for large-scale autonomous driving applications. Code will be released.

تم تقديم إطار جديد يسمى IDSplat لإعادة بناء مشاهد القيادة الديناميكية، باستخدام نهج التشتت الغاوسي ثلاثي الأبعاد الذاتي الإشراف الذي يسمح بتفكيك الكائنات بشكل صريح ومسارات حركة قابلة للتعلم دون الحاجة إلى تعليقات بشرية. تعالج هذه الطريقة التحديات التي واجهتها الأساليب السابقة التي اعتمدت على تعليقات مكلفة أو افتقرت إلى فصل واضح على مستوى الكائنات.

Se ha introducido un nuevo marco llamado IDSplat para reconstruir escenas de conducción dinámicas, utilizando un enfoque de esparcimiento gaussiano 3D auto-supervisado que permite una descomposición de instancias explícita y trayectorias de movimiento aprendibles sin necesidad de anotaciones humanas. Este método aborda los desafíos que enfrentaban las técnicas anteriores que dependían de anotaciones costosas o carecían de una separación clara a nivel de objeto.

Un nouveau cadre appelé IDSplat a été introduit pour reconstruire des scènes de conduite dynamiques, utilisant une approche d'éclatement gaussien 3D auto-supervisée qui permet une décomposition d'instance explicite et des trajectoires de mouvement apprenables sans nécessiter d'annotations humaines. Cette méthode répond aux défis rencontrés par les techniques précédentes qui dépendaient d'annotations coûteuses ou manquaient de séparation claire au niveau des objets.

A new framework called IDSplat has been introduced for reconstructing dynamic driving scenes, utilizing a self-supervised 3D Gaussian Splatting approach that allows for explicit instance decomposition and learnable motion trajectories without the need for human annotations. This method addresses the challenges faced by previous techniques that relied on costly annotations or lacked clear object-level separation.

IDSplat: Instance-Decomposed 3D Gaussian Splatting for Driving Scenes

One More Thing in AI – Your Shortcut to AI Mastery

IDSplat: Instance-Decomposed 3D Gaussian Splatting for Driving Scenes

Was this article worth reading? Share it

One More Thing in AI

LucidQuery AI

Attentive AI

Mapfit

Deptho.ai

Open Source Surveillance

Ready to build your own newsroom?