Summary
FL4N is presented as the first feed-forward latent model for 4D novel view synthesis (NVS), addressing limitations of monocular feed-forward methods that rely on explicit 4D Gaussian representations, which often suffer from temporal inconsistency and rendering artifacts. Instead of explicit 4D primitives, FL4N encodes a 4D scene prior into a unified latent representation spanning space and time. Its core insight is that modeling temporal dynamics as residuals on top of a pretrained static latent prior enables effective 4D learning even with limited dynamic training data. A lightweight dynamic encoder leverages static scene features to disentangle object motion from camera-induced parallax. Trained purely with an image synthesis objective, FL4N achieves state-of-the-art 4D NVS quality with real-time rendering, generalizes to real-world videos, and disentangles viewpoint from time. Paper: arXiv 2610.35854.
Paper Overview
- Field: Computer Vision
- Authors: Eunbeen Hong, Hyeonseo Yu, Honggyu An, Jooho Son, Hyunah Ko et al.
- Published: 2026-10-26
- arXiv: 2610.35854
Abstract
4D novel view synthesis (NVS) aims to render dynamic scenes across viewpoints and timestamps, which requires jointly inferring 3D structure and temporal dynamics. Existing monocular feed-forward methods often rely on explicit 4D Gaussian representations, which are prone to temporal inconsistency and severe rendering artifacts. We present FL4N, the first feed-forward latent 4D NVS model. Instead of relying on explicit 4D primitives, FL4N encodes a 4D scene prior into a unified latent representation across space and time. Our core insight is that learning temporal dynamics as residuals on top of a pretrained static latent prior enables effective 4D modeling with limited dynamic training data. To this end, we introduce a lightweight dynamic encoder that leverages static scene features to disentangle object motion from camera-induced parallax. Trained solely with an image synthesis objective, FL4N achieves state-of-the-art 4D NVS quality while rendering in real time. Its learned latent representation generalizes to real-world videos and disentangles viewpoint from time.Key Takeaways
- First feed-forward latent-space model for 4D novel view synthesis, avoiding explicit 4D primitives such as 4D Gaussians.
- Learns temporal dynamics as residuals over a pretrained static latent prior, enabling effective 4D modeling with limited dynamic training data.
- A lightweight dynamic encoder disentangles object motion from camera-induced parallax using static scene features.
- Trained only with an image synthesis objective; achieves state-of-the-art quality with real-time rendering.
- The learned latent representation generalizes to real-world videos and disentangles viewpoint from time.
---
*Auto-collected on 2026-10-28*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178636533