4D World Models: Bridging Generation and Reconstruction

📋 About the Workshop

This workshop focuses on 4D reconstruction and generation, bridging the gap between reconstructing and generating dynamic virtual worlds. The workshop brings together researchers working on neural representations, dynamic scene understanding, and efficient rendering for AR/VR and robotics applications.

Workshop Topics:

Generative and Editable World: Text/image-driven 3D synthesis, structured representations, and semantic editing for interactive modeling

Efficient and Scalable World Rendering: Real-time rendering, feed-forward reconstruction, and efficient Gaussian Splatting for AR/VR robotics

Dynamic Scene Understanding and Data Capture: Spatio-temporal modeling with Radiance Fields, large-scale 4D datasets, and temporal coherence

Virtual Humans as Embodied Agents: Photo-realistic animatable avatars, immersive volumetric humans, and deformable interaction in worlds

Neural Representations: Neural Radiance Fields, 3D Gaussian Splatting, and hybrid methods for dynamic world models

Applications: XR experiences, robotics, autonomous driving, digital twins, and creative content creation

Broader Impact:

This workshop is highly relevant to the computer vision, graphics, and AI communities, as it unites several of the most active and rapidly evolving research directions: Radiance Fields, 3D Gaussian Splatting, dynamic scene reconstruction, and world models for perception and generation. These threads converge on a shared ambition to endow AI systems with the capacity to understand, reconstruct, and generate coherent representations of dynamic world. By bridging the traditionally distinct domains of reconstruction and generation, the workshop promotes a unified view of scene modeling that directly supports progress in embodied intelligence, simulation, and creative content creation.

📢 Call for Papers

We invite submissions on:

4D Reconstruction and Generation: Methods for reconstructing or generating dynamic 3D scenes from multi-view videos, single videos, or text/image prompts
Neural Representations: Novel neural scene representations including NeRFs, 3D Gaussian Splatting, and hybrid approaches for dynamic scenes
Temporal Coherence: Techniques for maintaining consistency across time in dynamic 3D reconstructions and generations
Efficient Rendering: Real-time and low-latency rendering methods for 4D content in AR/VR and robotics
Human Avatars: Photo-realistic, animatable avatars and human performance capture
Applications: Novel applications in XR, robotics, autonomous driving, digital twins, and creative industries
Datasets and Benchmarks: New datasets, evaluation metrics, and benchmarks for 4D vision tasks
Other Relevant Topics: We also welcome submissions on related topics not explicitly listed above.

Submission Types:

Full Papers: Novel research contributions (up to 8 pages + references).
Dataset Papers: Novel datasets with benchmarks and baselines (up to 8 pages + references).

Important Dates:

Submission Opens: February 20, 2026
Submission Deadline: March 17, 2026 (11:59 PM AoE)
Author Notification: March 31, 2026
Camera-Ready Deadline: April 7, 2026
Workshop Date: CVPR 2026 (June 2026)

Submission Portal: Papers will be reviewed through CMT (link will be available shortly).

All submissions will undergo peer review. Accepted papers will be presented as spotlight talks at the workshop and will be published in the CVPR 2026 proceedings.