4DAnyone: 4D Human Reconstruction from Casual Monocular Video

Forum topic · 小凯 · 2026-08-22

Summary

4DAnyone is a computer vision framework that reconstructs 4D (spatio-temporal) humans from uncalibrated monocular video by generating reconstruction-level multi-view consistent videos and lifting them into 4D Gaussian Splatting (4DGS). The paper identifies a fundamental limitation of existing camera-control video diffusion models: when the number of target views required for 4DGS reconstruction exceeds the capacity of a single DiT forward pass, views must be processed in groups, causing two coupled bottlenecks. On the reference-context side, conditioning grows with already generated views, weakening cross-view appearance guidance; on the target-context side, disjoint groups cannot exchange information directly, leading to global structural drift. The authors call this the bounded attention context problem. 4DAnyone introduces two complementary solutions: Reference Context Packing, which compresses growing reference views into a fixed-length mixed-resolution context, and Target Context Routing, which rotates target view groupings during denoising. Experiments on DNA-Rendering and DyMVHumans benchmarks show that 4DAnyone outperforms prior methods in both novel-view video quality and downstream 4DGS reconstruction. The work is available on arXiv (2608.20335).

4DAnyone: Create Anyone in 4D from a Casual Monocular Video

Research area: Computer Vision (CV)
Authors: Yudong Jin, Tao Xie, Qihang Zhang, Zehong Shen
Published: 2026-08-22
arXiv: 2608.20335

Overview

4DAnyone is a framework for reconstructing 4D humans from uncalibrated monocular video. It works by generating reconstruction-level, multi-view consistent videos and lifting them into 4D Gaussian Splatting (4DGS).

Problem: The Bounded Attention Context Issue

Existing camera-control video diffusion models can synthesize novel-view videos, but consistency breaks down when the number of target views reaches the dozens required for 4DGS reconstruction. When target views exceed the capacity of a single DiT forward pass, they must be split into groups, producing two coupled bottlenecks:

Reference context side: the conditioning grows with already-generated views, weakening cross-view appearance guidance.
Target context side: disjoint groups cannot directly exchange information, causing global structural drift.

The authors identify this as the *bounded attention context* problem.

Method

4DAnyone addresses these bottlenecks with two complementary designs:

1. Reference Context Packing — compresses the growing set of reference views into a fixed-length, mixed-resolution context. 2. Target Context Routing — rotates the grouping of target views during the denoising process, enabling information exchange across groups.

Results

Experiments on the DNA-Rendering and DyMVHumans benchmarks show that 4DAnyone outperforms previous methods in both:

Novel-view video quality
Downstream 4DGS reconstruction quality

---

*Auto-collected on 2026-08-22.*