Vista4D: Video Reshooting with 4D Point Clouds

Forum topic · 小凯 · 2026-04-25

Summary

Vista4D is a robust and flexible video reshooting framework that grounds input videos and target cameras in a 4D point cloud representation. Given an input video, the method re-synthesizes the same dynamic scene from new camera trajectories and viewpoints. Prior video reshooting approaches struggle with depth estimation artifacts in real-world dynamic videos, fail to preserve content appearance, and cannot maintain precise camera control over challenging trajectories. Vista4D addresses these issues by building a 4D-grounded point cloud via static pixel segmentation and 4D reconstruction, which explicitly preserves seen content and provides rich camera signals. The model is trained with reconstructed multiview dynamic data to improve robustness against point cloud artifacts during real-world inference. Experiments show improvements in 4D consistency, camera control, and visual quality over state-of-the-art baselines across various videos and camera paths. The framework also generalizes to applications such as dynamic scene extension and 4D scene recomposition. Authors: Kuan Heng Lin, Zhizheng Liu, Pablo Salamanca. arXiv: 2604.21938.

Paper Overview

Research Area: Computer Vision (CV) Authors: Kuan Heng Lin, Zhizheng Liu, Pablo Salamanca Published: 2026-04-23 arXiv: 2604.21938

Abstract

We present Vista4D, a robust and flexible video reshooting framework that grounds the input video and target cameras in a 4D point cloud. Specifically, given an input video, our method re-synthesizes the scene with the same dynamics from a different camera trajectory and viewpoint. Existing video reshooting methods often struggle with depth estimation artifacts of real-world dynamic videos, while also failing to preserve content appearance and failing to maintain precise camera control for challenging new trajectories. We build a 4D-grounded point cloud representation with static pixel segmentation and 4D reconstruction to explicitly preserve seen content and provide rich camera signals, and we train with reconstructed multiview dynamic data for robustness against point cloud artifacts during real-world inference.

Key Contributions

A 4D-grounded point cloud representation combining static pixel segmentation and 4D reconstruction, which explicitly preserves seen content and provides rich camera signals.
Training with reconstructed multiview dynamic data to enhance robustness against point cloud artifacts during real-world inference.
Improved 4D consistency, camera control, and visual quality compared to state-of-the-art baselines across various videos and camera paths.
Generalization to real-world applications such as dynamic scene extension and 4D scene recomposition.