Kun Xu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Kun Xu has a total of 74 co-authors. Here are the top co-authors by number of shared publications:
- Ling-Qi Yan
Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi, United Arab Emirates
5 shared publications - Shaokun Zheng
Tsinghua University, Beijing, China
4 shared publications - Xuejun Hu
Tsinghua University, Beijing, China
4 shared publications - Jianchao Tan
Kuaishou Technology, Beijing, China
3 shared publications - Jiaqi Wu
CS Dept, Tsinghua University, BEIJING, China
3 shared publications Ralph R. Martin
School of Computer Science and Informatics, Cardiff University, Cardiff, U.K
3 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Graphics
- Vol. 45 (6) , 1-14
Pixel-Light Attention for Neural Many-Light Importance Sampling
We present a neural approach for many-light importance sampling built on a pixel-light attention mechanism for guided hierarchical traversal. Prior learning-based methods confine themselves to a single coarse level of the light hierarchy and entangle pixel and light information in a single learned object, limiting how finely the sampling distribution can adapt. Our attention scores each branch via a dot product between a learnable pixel feature and a direction-aware light node feature based on spherical-harmonic bases — factorizing the two sides into separate learnable representations while remaining lightweight enough to run at every traversal step. We train all features online during rendering by minimizing a KL-divergence loss against the ideal distribution, with gradients accumulated along each path via an explicit re-traversal pass. Our method achieves consistently lower error than state-of-the-art techniques across a range of challenging scenes.
- 9 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Graphics
- Vol. 45 (6) , 1-14
Open AccessPixel-Light Attention for Neural Many-Light Importance Sampling
We present a neural approach for many-light importance sampling built on a pixel-light attention mechanism for guided hierarchical traversal. Prior learning-based methods confine themselves to a single coarse level of the light hierarchy and entangle pixel and light information in a single learned object, limiting how finely the sampling distribution can adapt. Our attention scores each branch via a dot product between a learnable pixel feature and a direction-aware light node feature based on spherical-harmonic bases — factorizing the two sides into separate learnable representations while remaining lightweight enough to run at every traversal step. We train all features online during rendering by minimizing a KL-divergence loss against the ideal distribution, with gradients accumulated along each path via an explicit re-traversal pass. Our method achieves consistently lower error than state-of-the-art techniques across a range of challenging scenes.
- Open Access
- 19 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-11
Spatio-Temporal Control Variates with ReSTIR for Real-Time Rendering
- 19 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-11
Open AccessSpatio-Temporal Control Variates with ReSTIR for Real-Time Rendering
- Open Access
- 3 July 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Graphics
- Vol. 45 (4) , 1-18
Generalized Spherical Harmonics Products using Spherical Grids
Spherical Harmonics (SH) are fundamental mathematical tools for compactly representing low-frequency spherical functions in computer graphics. Evaluating products of SH-represented functions is a core operation in many rendering algorithms and often becomes a major computational bottleneck. Existing SH product methods face two key challenges: insufficient computational efficiency, and the inflexible requirement that all input functions share the same SH order. To overcome these limitations, we introduce the Generalized Spherical Harmonics Product , a novel formulation that supports inputs with different SH orders and allows flexible truncation of the output. We further propose an efficient and practical computation approach based on Spherical Grids (SphGrid). We establish explicit bidirectional conversions between SphGrid representations and SH coefficients, thus enabling SphGrid to serve as an intermediate representation. Under this framework, SH product computation is reduced to simple point-wise multiplications on the grid. We then derive sufficient conditions on the grid resolution to guarantee exact computation. Using lower resolutions yields an approximate variant that provides a favorable trade-off between accuracy and efficiency. Our approach is not only more general but also demonstrates superior performance even for the specialized case of identical input orders. In particular, our accurate method achieves a 3.5–6.0× speedup over the state-of-the-art method in a practical setting for graphics, while our approximate variant yields lower errors and a 2.5–6.5× speedup over the state-of-the-art approximate method. We rigorously analyze the correctness and efficiency of our method and demonstrate its effectiveness in real-time rendering applications.
- 3 July 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Graphics
- Vol. 45 (4) , 1-18
Open AccessGeneralized Spherical Harmonics Products using Spherical Grids
Spherical Harmonics (SH) are fundamental mathematical tools for compactly representing low-frequency spherical functions in computer graphics. Evaluating products of SH-represented functions is a core operation in many rendering algorithms and often becomes a major computational bottleneck. Existing SH product methods face two key challenges: insufficient computational efficiency, and the inflexible requirement that all input functions share the same SH order. To overcome these limitations, we introduce the Generalized Spherical Harmonics Product , a novel formulation that supports inputs with different SH orders and allows flexible truncation of the output. We further propose an efficient and practical computation approach based on Spherical Grids (SphGrid). We establish explicit bidirectional conversions between SphGrid representations and SH coefficients, thus enabling SphGrid to serve as an intermediate representation. Under this framework, SH product computation is reduced to simple point-wise multiplications on the grid. We then derive sufficient conditions on the grid resolution to guarantee exact computation. Using lower resolutions yields an approximate variant that provides a favorable trade-off between accuracy and efficiency. Our approach is not only more general but also demonstrates superior performance even for the specialized case of identical input orders. In particular, our accurate method achieves a 3.5–6.0× speedup over the state-of-the-art method in a practical setting for graphics, while our approximate variant yields lower errors and a 2.5–6.5× speedup over the state-of-the-art approximate method. We rigorously analyze the correctness and efficiency of our method and demonstrate its effectiveness in real-time rendering applications.
- Open Access
- 3 July 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Graphics
- Vol. 45 (4) , 1-11
Gradient Domain Reconstruction for Monte Carlo PDE Solvers
Grid-free Monte Carlo methods are capable of solving Poisson equations on highly complex domains. However, existing methods operate solely in the primal domain and can converge slowly due to high variance. Inspired by gradient-domain rendering, we introduce a gradient-domain framework for Poisson problems. Specifically, we devise a new Monte Carlo estimator that directly targets differences of the solution between spatially varying query locations. Further, we adopt state-of-the-art reconstruction techniques originated in gradient-domain rendering to allow efficient reconstruction of the solutions without incurring additional bias. We demonstrate the effectiveness of our technique by comparing solutions obtained using our method and several state-of-the-art baselines.
- 3 July 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Graphics
- Vol. 45 (4) , 1-11
Open AccessGradient Domain Reconstruction for Monte Carlo PDE Solvers
Grid-free Monte Carlo methods are capable of solving Poisson equations on highly complex domains. However, existing methods operate solely in the primal domain and can converge slowly due to high variance. Inspired by gradient-domain rendering, we introduce a gradient-domain framework for Poisson problems. Specifically, we devise a new Monte Carlo estimator that directly targets differences of the solution between spatially varying query locations. Further, we adopt state-of-the-art reconstruction techniques originated in gradient-domain rendering to allow efficient reconstruction of the solutions without incurring additional bias. We demonstrate the effectiveness of our technique by comparing solutions obtained using our method and several state-of-the-art baselines.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Visualization and Computer Graphics
- Vol. 32 (10) , 1-13
Blue Noise Dithering for Reservoir-based Spatio-temporal Importance Resampling
Monte Carlo path tracing remains the standard approach for photorealistic image synthesis, but its high variance makes efficient rendering challenging. Reservoir-based Spatiotemporal Importance Resampling (ReSTIR) has recently emerged as a promising framework for efficiently reusing samples in real-time path tracing. However, the stochastic resampling step of ReSTIR makes it incompatible with ideal sample distributions like blue-noise. In this work, we reinterpret the ReSTIR process through the lens of dithered sampling and establish a connection between reservoir resampling and ordered blue-noise sampling. Building on this perspective, we propose a blue-noise-driven spatiotemporal resampling scheme that preserves perceptually favorable high-frequency error distribution during reuse. We provide a full integration of the proposed method into a real-time path tracer. Our experiments demonstrate that integrating blue noise into ReSTIR can improve the perceptual quality of low-sample-count rendered images across a range of scenes. Quantitative evaluations show reductions in perceptual error metrics with only marginal runtime overhead, maintaining efficiency suitable for real-time rendering applications.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Visualization and Computer Graphics
- Vol. 32 (10) , 1-13
Blue Noise Dithering for Reservoir-based Spatio-temporal Importance Resampling
Monte Carlo path tracing remains the standard approach for photorealistic image synthesis, but its high variance makes efficient rendering challenging. Reservoir-based Spatiotemporal Importance Resampling (ReSTIR) has recently emerged as a promising framework for efficiently reusing samples in real-time path tracing. However, the stochastic resampling step of ReSTIR makes it incompatible with ideal sample distributions like blue-noise. In this work, we reinterpret the ReSTIR process through the lens of dithered sampling and establish a connection between reservoir resampling and ordered blue-noise sampling. Building on this perspective, we propose a blue-noise-driven spatiotemporal resampling scheme that preserves perceptually favorable high-frequency error distribution during reuse. We provide a full integration of the proposed method into a real-time path tracer. Our experiments demonstrate that integrating blue noise into ReSTIR can improve the perceptual quality of low-sample-count rendered images across a range of scenes. Quantitative evaluations show reductions in perceptual error metrics with only marginal runtime overhead, maintaining efficiency suitable for real-time rendering applications.
- Open Access
- 14 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-2
Topology-Aware Polymorphism for Embedded Shading Languages
- 14 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-2
Open AccessTopology-Aware Polymorphism for Embedded Shading Languages
- Open Access
- 4 December 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Graphics
- Vol. 44 (6) , 1-14
Generalized Unbiased Reconstruction for Gradient-Domain Rendering
Gradient-domain rendering estimates image-space gradients using correlated sampling, which can be combined with color information to reconstruct smoother and less noisy images. While simple ℒ 2 reconstruction is unbiased, it often leads to visible artifacts. In contrast, most recent reconstruction methods based on learned or handcrafted techniques improve visual quality but introduce bias, leaving the development of practically unbiased reconstruction approaches relatively underexplored. In this work, we propose a generalized framework for unbiased reconstruction in gradient-domain rendering. We first derive the unbiasedness condition under a general formulation that linearly combines pixel colors and gradients. Based on this unbiasedness condition, we design a practical algorithm 1 that minimizes image variance while strictly satisfying unbiasedness. Experimental results demonstrate that our method not only guarantees unbiasedness but also achieves superior quality compared to existing unbiased and slightly biased reconstruction methods.
- 4 December 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Graphics
- Vol. 44 (6) , 1-14
Open AccessGeneralized Unbiased Reconstruction for Gradient-Domain Rendering
Gradient-domain rendering estimates image-space gradients using correlated sampling, which can be combined with color information to reconstruct smoother and less noisy images. While simple ℒ 2 reconstruction is unbiased, it often leads to visible artifacts. In contrast, most recent reconstruction methods based on learned or handcrafted techniques improve visual quality but introduce bias, leaving the development of practically unbiased reconstruction approaches relatively underexplored. In this work, we propose a generalized framework for unbiased reconstruction in gradient-domain rendering. We first derive the unbiasedness condition under a general formulation that linearly combines pixel colors and gradients. Based on this unbiasedness condition, we design a practical algorithm 1 that minimizes image variance while strictly satisfying unbiasedness. Experimental results demonstrate that our method not only guarantees unbiasedness but also achieves superior quality compared to existing unbiased and slightly biased reconstruction methods.
- 16 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 34 (10577149) , 5920-5934
Fast Video Recoloring via Curve-Based Palettes
Color grading, as a crucial step in film post-production, plays an important role in emotional expression and artistic enhancement. Recently, a geometric palette-based approach to video recoloring has been introduced with impressive results. It offers an intuitive interface that allows users to alter the color of a video by manipulating a limited set of representative colors. However, this method has two primary limitations. Firstly, palette extraction is computationally expensive, often taking more than one hour to generate palettes even for medium-length videos, which significantly limits the practical application of color editing for longer videos. Secondly, the palette colors are less representative, and some primary colors may be omitted from the resulting palettes during topological simplification, making it less intuitive in color editing. To overcome these limitations, in this paper, we propose a novel approach to video recoloring. The core of our method is a set of Bézier curves that connect the dominant colors throughout the input video. By slicing these Bézier curves in RGBT space, per-frame palette can be naturally derived. During recoloring, users can select several frames of interest and modify their corresponding palettes to change the color of the video. Our method is simple and intuitive, enabling compelling time-varying recoloring results. Compared to existing methods, our approach is more efficient in palette extraction and can effectively capture the dominant colors of the video. Extensive experiments demonstrate the effectiveness of our method.
- 16 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 34 (10577149) , 5920-5934
Fast Video Recoloring via Curve-Based Palettes
Color grading, as a crucial step in film post-production, plays an important role in emotional expression and artistic enhancement. Recently, a geometric palette-based approach to video recoloring has been introduced with impressive results. It offers an intuitive interface that allows users to alter the color of a video by manipulating a limited set of representative colors. However, this method has two primary limitations. Firstly, palette extraction is computationally expensive, often taking more than one hour to generate palettes even for medium-length videos, which significantly limits the practical application of color editing for longer videos. Secondly, the palette colors are less representative, and some primary colors may be omitted from the resulting palettes during topological simplification, making it less intuitive in color editing. To overcome these limitations, in this paper, we propose a novel approach to video recoloring. The core of our method is a set of Bézier curves that connect the dominant colors throughout the input video. By slicing these Bézier curves in RGBT space, per-frame palette can be naturally derived. During recoloring, users can select several frames of interest and modify their corresponding palettes to change the color of the video. Our method is simple and intuitive, enabling compelling time-varying recoloring results. Compared to existing methods, our approach is more efficient in palette extraction and can effectively capture the dominant colors of the video. Extensive experiments demonstrate the effectiveness of our method.
- Open Access
- 9 August 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-2
Adding Regional Control for Continuous Remeshing via Attention Flows
- 9 August 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-2
Open AccessAdding Regional Control for Continuous Remeshing via Attention Flows
- Open Access
- 27 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-10
Kernel Predicting Neural Shadow Maps
- 27 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-10
Open AccessKernel Predicting Neural Shadow Maps