OpenNovelty.org

Dynamic Texture Modeling of 3D Clothed Gaussian Avatars from a Single Video

ICLR 2026 Conference SubmissionAnonymous Authors
OpenReview Score: 6.5Download Report PDF
3D Computer VisionNeural Rendering3D Avatar Modeling
Abstract:

Recent advances in neural rendering, particularly 3D Gaussian Splatting (3DGS), have enabled animatable 3D human avatars from single videos with efficient rendering and high fidelity. However, current methods struggle with dynamic appearances, especially in loose garments (e.g., skirts), causing unrealistic cloth motion and needle artifacts. This paper introduces a novel approach to dynamic appearance modeling for 3DGS-based avatars, focusing on loose clothing. We identify two key challenges: (1) limited Gaussian deformation under pre-defined template articulation, and (2) a mismatch between body-template assumptions and the geometry of loose apparel. To address these issues, we propose a motion-aware autoregressive structural deformation framework for Gaussians. We structure Gaussians into an approximate graph and recursively predict structure-preserving updates, yielding realistic, template-free cloth dynamics. Our framework enables view-consistent and robust appearance modeling under the single-view constraint, producing accurate foreground silhouettes and precise alignment of Gaussian points with clothed shapes. To demonstrate the effectiveness of our method, we introduce an in-the-wild dataset featuring subjects performing dynamic movements in loose clothing, and extensive experiments validate that our approach significantly outperforms existing 3DGS-based methods in modeling dynamic appearances from single videos.

Disclaimer
This report is AI-GENERATED using Large Language Models and WisPaper (A scholar search engine). It analyzes academic papers' tasks and contributions against retrieved prior work. While this system identifies POTENTIAL overlaps and novel directions, ITS COVERAGE IS NOT EXHAUSTIVE AND JUDGMENTS ARE APPROXIMATE. These results are intended to assist human reviewers and SHOULD NOT be relied upon as a definitive verdict on novelty.
NOTE that some papers exist in multiple, slightly different versions (e.g., with different titles or URLs). The system may retrieve several versions of the same underlying work. The current automated pipeline does not reliably align or distinguish these cases, so human reviewers will need to disambiguate them manually.
If you have any questions, please contact: mingzhang23@m.fudan.edu.cn

Overview

Overall Novelty Assessment

The paper proposes a motion-aware autoregressive structural deformation framework for modeling loose clothing in 3D Gaussian Splatting avatars. It occupies the 'Autoregressive and Graph-Based Deformation' leaf within the 'Motion-Aware Structural Deformation Frameworks' branch. Notably, this leaf contains only the original paper itself—no sibling papers are listed—indicating a relatively sparse research direction. The taxonomy shows 15 total papers across 11 leaf nodes, suggesting the field is moderately populated but with uneven distribution across approaches.

The taxonomy reveals neighboring directions that contextualize this work. The sibling leaf 'Disentangled Motion and Appearance Modeling' contains two papers (PGC, Topology Aware Gaussian) that also address non-rigid dynamics but emphasize explicit separation of motion tracking from appearance. Adjacent branches include 'Physics-Based Cloth Simulation' (three leaves, six papers total) focusing on simulation engines and material properties, and 'Canonical Space Parameterization' (two leaves, three papers) using UV-space representations. The paper's graph-structured autoregressive approach diverges from both physics-driven simulation and static canonical mappings, positioning it between learned temporal modeling and geometry-aware deformation.

Among 30 candidates examined, the contribution-level analysis shows varied novelty profiles. The core 'motion-aware autoregressive structural deformation framework' examined 10 candidates with zero refutable matches, suggesting limited direct prior work in this specific formulation. Similarly, the 'dynamic appearance modeling approach for loose clothing' found no refutations among 10 candidates. However, the 'in-the-wild dataset' contribution encountered one refutable candidate among 10 examined, indicating some overlap in dataset construction. The limited search scope (30 papers, not exhaustive) means these statistics reflect top-K semantic matches rather than comprehensive field coverage.

Based on the top-30 semantic search results, the work appears to occupy a relatively novel position within motion-aware deformation frameworks, particularly in combining autoregressive prediction with graph-structured Gaussians. The sparse population of its taxonomy leaf and low refutation rates across core contributions suggest incremental advancement over existing methods, though the analysis cannot rule out relevant work outside the examined candidate set. The dataset contribution shows more substantial prior overlap, consistent with established practices in avatar reconstruction benchmarks.

Taxonomy

Core-task Taxonomy Papers
15
3
Claimed Contributions
30
Contribution Candidate Papers Compared
1
Refutable Paper

Research Landscape Overview

Core task: Dynamic appearance modeling of loose clothing in 3D Gaussian avatars. The field addresses the challenge of reconstructing and animating realistic human avatars with clothing that exhibits complex, motion-dependent deformations. The taxonomy reveals several complementary strategies: Physics-Based Cloth Simulation and Dynamics leverages physical constraints and material properties to drive garment motion (e.g., PhysAvatar[1], PBDyG[4]); Motion-Aware Structural Deformation Frameworks learn data-driven mappings between body pose and garment shape, often using graph or autoregressive architectures; Canonical Space Parameterization and UV Mapping establishes stable texture coordinates for tracking appearance across frames (e.g., Learning UV Loose Clothing[14]); Layered and Disentangled Garment Representations separate body and clothing into distinct geometric or neural components (e.g., ClotheDreamer[3], Garment Aware Gaussian[12]); and Expressive Full-Body Avatar Reconstruction focuses on capturing fine-grained details and diverse body motions in holistic systems (e.g., RealityAvatar[2], EVA[7]). A central tension runs between physics-driven and learning-based approaches: physics simulators offer interpretability and generalization to unseen motions but can be computationally expensive, while neural deformation models achieve real-time performance yet may struggle with extreme poses. Within Motion-Aware Structural Deformation Frameworks, Dynamic Texture Clothed Gaussian[0] emphasizes autoregressive and graph-based deformation to capture temporal dependencies in loose garments, positioning itself alongside works like PGC[5] and Topology Aware Gaussian[6] that also exploit mesh connectivity for coherent shape updates. Compared to purely canonical-space methods such as Learning UV Loose Clothing[14], Dynamic Texture Clothed Gaussian[0] integrates motion history more explicitly, while differing from physics-heavy pipelines like PhysAvatar[1] by prioritizing learned priors over explicit simulation. Open questions remain around balancing physical plausibility with rendering speed, handling topology changes during extreme deformation, and generalizing learned models to novel garment types.

Claimed Contributions

Motion-aware autoregressive structural deformation framework for Gaussians

The authors introduce a framework that structures 3D Gaussians into a graph and recursively predicts deformations to model realistic, template-free cloth dynamics. This approach addresses temporal context-unaware Gaussian deformation by incorporating motion awareness and autoregressive prediction.

10 retrieved papers
Dynamic appearance modeling approach for loose clothing in 3DGS-based avatars

The authors present a new method specifically designed to handle dynamic appearances in loose garments for 3D Gaussian Splatting avatars. This addresses limitations in existing methods that struggle with realistic cloth motion and needle artifacts in loose clothing.

10 retrieved papers
In-the-wild dataset with subjects in loose clothing performing dynamic movements

The authors contribute a new dataset containing subjects wearing loose clothing and performing dynamic movements. This dataset enables evaluation of methods for modeling secondary motion and dynamic appearances in challenging real-world scenarios.

10 retrieved papers
Can Refute

Core Task Comparisons

Comparisons with papers in the same taxonomy category

Within the taxonomy built over the current TopK core-task papers, the original paper is assigned to a leaf with no direct siblings and no cousin branches under the same grandparent topic. In this retrieved landscape, it appears structurally isolated, which is one partial signal of novelty, but still constrained by search coverage and taxonomy granularity.

Contribution Analysis

Detailed comparisons for each claimed contribution

Contribution

Motion-aware autoregressive structural deformation framework for Gaussians

The authors introduce a framework that structures 3D Gaussians into a graph and recursively predicts deformations to model realistic, template-free cloth dynamics. This approach addresses temporal context-unaware Gaussian deformation by incorporating motion awareness and autoregressive prediction.

Contribution

Dynamic appearance modeling approach for loose clothing in 3DGS-based avatars

The authors present a new method specifically designed to handle dynamic appearances in loose garments for 3D Gaussian Splatting avatars. This addresses limitations in existing methods that struggle with realistic cloth motion and needle artifacts in loose clothing.

Contribution

In-the-wild dataset with subjects in loose clothing performing dynamic movements

The authors contribute a new dataset containing subjects wearing loose clothing and performing dynamic movements. This dataset enables evaluation of methods for modeling secondary motion and dynamic appearances in challenging real-world scenarios.

Table of Contents