Xiaowei Zhou
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Xiaowei Zhou has a total of 196 co-authors. Here are the top co-authors by number of shared publications:
- Hujun Bao
State Key Lab of CAD&CG, Zhejiang University, Hangzhou, China
29 shared publications - Sida Peng
Zhejiang University, Hangzhou, China
23 shared publications - Jiaming Sun
DV.ai, Hangzhou, China
11 shared publications - Qing Shuai
Zhejiang University, Hangzhou, China
8 shared publications - Haotong Lin
Zhejiang University, Hangzhou, China
6 shared publications - Junting Dong
Shanghai Artificial Intelligence Laboratory, Shanghai, China
6 shared publications
Latest Publications
- Open Access
- 30 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Volumetric Video Challenge 2026
- 30 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Open AccessVolumetric Video Challenge 2026
- Open Access
- 30 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-11
4DAnyone: Create Anyone in 4D from a Casual Monocular Video
- 30 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-11
Open Access4DAnyone: Create Anyone in 4D from a Casual Monocular Video
- Open Access
- 30 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-12
RealMe: Immersive Volumetric Video from Minimal Captures
- 30 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-12
Open AccessRealMe: Immersive Volumetric Video from Minimal Captures
- Open Access
- 30 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-11
Track4D-Calib: Track-Driven Geometric Closure for 4D Camera-Array Calibration
- 30 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-11
Open AccessTrack4D-Calib: Track-Driven Geometric Closure for 4D Camera-Array Calibration
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Graphics
- Vol. 45 (6) , 1-15
InfiniSplat: Implicit Gaussian Decoding for Large-Baseline Monocular View Synthesis
Single-image feed-forward 3D Gaussian Splatting (3DGS) aims to directly generate a renderable 3D scene representation from one input image, avoiding the cost of multi-view capture and per-scene optimization. However, existing methods are often constrained by a pixel-aligned representation, where Gaussians are predicted from fixed image-grid locations. Such pixel-aligned primitives can produce promising nearby-view renderings, but they remain weakly coupled to underlying scene surfaces and struggle to preserve coherent structures under large viewpoint shifts. We present InfiniSplat , a feed-forward single-image 3DGS framework that moves from a pixel-aligned representation toward a surface-aligned representation. InfiniSplat constructs this representation by first using geometry-guided sampling to place 2D supports according to depth-induced local surface structure, and then applying a query-conditioned implicit decoder to predict Gaussian attributes from the image features queried at these supports. By grounding support locations in geometry while decoupling Gaussian prediction from fixed pixel centers, InfiniSplat produces Gaussian layouts that better follow scene surfaces and reduce scattered primitives caused by grid discretization. Across multiple cross-dataset NVS evaluations, InfiniSplat achieves state-of-the-art performance compared with single-image feedforward baselines, and demonstrates zero-shot generalization from Hyper-sim indoor synthetic training to complex open-world scenes. Project page: https://zju3dv.github.io/InfiniSplat.
- 9 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Graphics
- Vol. 45 (6) , 1-15
Open AccessInfiniSplat: Implicit Gaussian Decoding for Large-Baseline Monocular View Synthesis
Single-image feed-forward 3D Gaussian Splatting (3DGS) aims to directly generate a renderable 3D scene representation from one input image, avoiding the cost of multi-view capture and per-scene optimization. However, existing methods are often constrained by a pixel-aligned representation, where Gaussians are predicted from fixed image-grid locations. Such pixel-aligned primitives can produce promising nearby-view renderings, but they remain weakly coupled to underlying scene surfaces and struggle to preserve coherent structures under large viewpoint shifts. We present InfiniSplat , a feed-forward single-image 3DGS framework that moves from a pixel-aligned representation toward a surface-aligned representation. InfiniSplat constructs this representation by first using geometry-guided sampling to place 2D supports according to depth-induced local surface structure, and then applying a query-conditioned implicit decoder to predict Gaussian attributes from the image features queried at these supports. By grounding support locations in geometry while decoupling Gaussian prediction from fixed pixel centers, InfiniSplat produces Gaussian layouts that better follow scene surfaces and reduce scattered primitives caused by grid discretization. Across multiple cross-dataset NVS evaluations, InfiniSplat achieves state-of-the-art performance compared with single-image feedforward baselines, and demonstrates zero-shot generalization from Hyper-sim indoor synthetic training to complex open-world scenes. Project page: https://zju3dv.github.io/InfiniSplat.
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Graphics
- Vol. 45 (6) , 1-17
DynOLAT: OLAT Video Prediction for Controllable Dynamic Full-Body Relighting
Accurate lighting response is essential for realistic dynamic human relighting, because shadows, highlights, and color changes should follow the target illumination in both direction and relative strength. Current learning-based relighting methods can generate visually plausible videos, but they often exhibit imprecise lighting responses: shadows and highlights may drift away from the target direction, and the output brightness may not faithfully follow the relative intensities in the target environment, making the result closer to appearance harmonization than controllable relighting. To address this issue, we propose DynOLAT, a generalizable OLAT video prediction framework for dynamic full-body humans. Instead of directly synthesizing a single relit appearance, our method predicts temporally consistent One-Light-At-a-Time (OLAT) basis videos from flat-lit inputs, and then linearly combines the predicted bases according to the target environment map. This OLAT-based formulation makes the directional response explicit in the predicted bases, while the relative lighting strength is determined by the linear weights derived from the environment map. We further introduce a dynamic capture pipeline and the first large-scale dynamic full-body OLAT dataset, comprising over 24 million frames from 224 identity-outfit combinations, where different outfits of the same person define separate combinations. Experiments show that DynOLAT responds more faithfully to variations in target lighting direction and relative intensity than the tested baselines, enabling controllable video relighting while preserving fine appearance details and temporal coherence. The dataset and processing tools will be publicly released to foster future research.
- 9 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Graphics
- Vol. 45 (6) , 1-17
Open AccessDynOLAT: OLAT Video Prediction for Controllable Dynamic Full-Body Relighting
Accurate lighting response is essential for realistic dynamic human relighting, because shadows, highlights, and color changes should follow the target illumination in both direction and relative strength. Current learning-based relighting methods can generate visually plausible videos, but they often exhibit imprecise lighting responses: shadows and highlights may drift away from the target direction, and the output brightness may not faithfully follow the relative intensities in the target environment, making the result closer to appearance harmonization than controllable relighting. To address this issue, we propose DynOLAT, a generalizable OLAT video prediction framework for dynamic full-body humans. Instead of directly synthesizing a single relit appearance, our method predicts temporally consistent One-Light-At-a-Time (OLAT) basis videos from flat-lit inputs, and then linearly combines the predicted bases according to the target environment map. This OLAT-based formulation makes the directional response explicit in the predicted bases, while the relative lighting strength is determined by the linear weights derived from the environment map. We further introduce a dynamic capture pipeline and the first large-scale dynamic full-body OLAT dataset, comprising over 24 million frames from 224 identity-outfit combinations, where different outfits of the same person define separate combinations. Experiments show that DynOLAT responds more faithfully to variations in target lighting direction and relative intensity than the tested baselines, enabling controllable video relighting while preserving fine appearance details and temporal coherence. The dataset and processing tools will be publicly released to foster future research.
- 13 September 2026
- Published by Springer Nature
- p. 306-323
Habitat-GS: A High-Fidelity Navigation Simulator with Dynamic Gaussian Splatting
- 13 September 2026
- Published by Springer Nature
- p. 306-323
Habitat-GS: A High-Fidelity Navigation Simulator with Dynamic Gaussian Splatting
- 12 September 2026
- Published by Springer Nature
- p. 615-632
PointSplat: Compact Gaussian Splatting via Human-Centric Prediction
- 12 September 2026
- Published by Springer Nature
- p. 615-632
PointSplat: Compact Gaussian Splatting via Human-Centric Prediction
- Open Access
3D Human Pose Estimation from Monocular Video Sequences in Underwater Scenarios
This paper presents a novel approach for estimating 3D human pose from monocular video sequences in underwater scenarios, tackling the unique challenges posed by water refraction, body occlusion, low image quality and illumination distortion in underwater environments. Leveraging both 2D keypoint extraction and parametric model estimation, our method operates in a two-stage framework including preprocessing and optimization. In the preprocessing stage, a Part Attention Regressor (PARE) is adopted to dynamically estimate SMPL human body parameters, particularly adept at handling occlusions common in underwater scenarios. Additionally, a 2D keypoint detector, employing YOLO for bounding box detection and HRNet for keypoint regression, enhances feature extraction despite underwater image challenges. In the optimization stage, we propose an underwater variational autoencoder (UW-VAE), which adopts a data-driven strategy to learn the biomechanical prior distribution of underwater human poses and implicitly correct unreasonable pose parameters caused by refraction and occlusion. The optimization process incorporates constraints aligning final SMPL models with detected 2D keypoints, minimizing disparity between adjusted and original SMPL models, and ensuring temporal consistency. Furthermore, to address the scarcity of annotated underwater datasets, we build a full pipeline to generate synthetic underwater datasets with complete annotations based on UW-VAE. Experimental results on the SwimXYZ synthetic dataset show that our method achieves 51.60% PCK@0.2 and 80.13% PCK@0.5, outperforming state-of-the-art land-based methods across most stroke categories. Validation on real-world underwater swimming datasets demonstrates improved 2D keypoint accuracy after synthetic-data fine-tuning, which provides a new solution for 3D human motion analysis in underwater sports, biomechanical research and swimming training.
Open Access3D Human Pose Estimation from Monocular Video Sequences in Underwater Scenarios
This paper presents a novel approach for estimating 3D human pose from monocular video sequences in underwater scenarios, tackling the unique challenges posed by water refraction, body occlusion, low image quality and illumination distortion in underwater environments. Leveraging both 2D keypoint extraction and parametric model estimation, our method operates in a two-stage framework including preprocessing and optimization. In the preprocessing stage, a Part Attention Regressor (PARE) is adopted to dynamically estimate SMPL human body parameters, particularly adept at handling occlusions common in underwater scenarios. Additionally, a 2D keypoint detector, employing YOLO for bounding box detection and HRNet for keypoint regression, enhances feature extraction despite underwater image challenges. In the optimization stage, we propose an underwater variational autoencoder (UW-VAE), which adopts a data-driven strategy to learn the biomechanical prior distribution of underwater human poses and implicitly correct unreasonable pose parameters caused by refraction and occlusion. The optimization process incorporates constraints aligning final SMPL models with detected 2D keypoints, minimizing disparity between adjusted and original SMPL models, and ensuring temporal consistency. Furthermore, to address the scarcity of annotated underwater datasets, we build a full pipeline to generate synthetic underwater datasets with complete annotations based on UW-VAE. Experimental results on the SwimXYZ synthetic dataset show that our method achieves 51.60% PCK@0.2 and 80.13% PCK@0.5, outperforming state-of-the-art land-based methods across most stroke categories. Validation on real-world underwater swimming datasets demonstrates improved 2D keypoint accuracy after synthetic-data fine-tuning, which provides a new solution for 3D human motion analysis in underwater sports, biomechanical research and swimming training.
- 3 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. PP (01628828) , 1-13
GVHMR: Gravity-View Coordinates for Global Human Motion Recovery From Monocular Videos
We present a novel method for recovering world-grounded human motion from monocular video. The main challenge lies in the ambiguity of defining the world coordinate system, which varies between sequences. Previous approaches attempt to alleviate this issue by predicting relative motion in an auto-regressive manner, but are prone to accumulating errors. Instead, we propose estimating human poses in a novel Gravity-View (GV) coordinate system, which is defined by the world gravity and the camera view direction. The proposed GV system is naturally gravity-aligned and uniquely defined for each video frame, largely reducing the ambiguity of learning image-pose mapping. The estimated poses can be transformed back to the world coordinate system using camera rotations, forming a global motion sequence. Additionally, the per-frame estimation avoids error accumulation in the auto-regressive methods. Experiments on in-the-wild benchmarks demonstrate that our method recovers more realistic motion in both the camera space and world-grounded settings, outperforming state-of-the-art methods in both accuracy and speed. We further introduce a Stationary Label Predictor (SLP) to estimate the motion stationary states of hands and feet. By producing sharper decision boundaries, the module mitigates unnatural artifacts such as foot-sliding and shows improved performance under fast-motion inputs. The code is available at https://zju3dv.github.io/gvhmr.
- 3 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. PP (01628828) , 1-13
GVHMR: Gravity-View Coordinates for Global Human Motion Recovery From Monocular Videos
We present a novel method for recovering world-grounded human motion from monocular video. The main challenge lies in the ambiguity of defining the world coordinate system, which varies between sequences. Previous approaches attempt to alleviate this issue by predicting relative motion in an auto-regressive manner, but are prone to accumulating errors. Instead, we propose estimating human poses in a novel Gravity-View (GV) coordinate system, which is defined by the world gravity and the camera view direction. The proposed GV system is naturally gravity-aligned and uniquely defined for each video frame, largely reducing the ambiguity of learning image-pose mapping. The estimated poses can be transformed back to the world coordinate system using camera rotations, forming a global motion sequence. Additionally, the per-frame estimation avoids error accumulation in the auto-regressive methods. Experiments on in-the-wild benchmarks demonstrate that our method recovers more realistic motion in both the camera space and world-grounded settings, outperforming state-of-the-art methods in both accuracy and speed. We further introduce a Stationary Label Predictor (SLP) to estimate the motion stationary states of hands and feet. By producing sharper decision boundaries, the module mitigates unnatural artifacts such as foot-sliding and shows improved performance under fast-motion inputs. The code is available at https://zju3dv.github.io/gvhmr.