Searcharxiv⌕ Search

arXiv subjects

Hejun Wang

Publications and source records attributed to Hejun Wang.

5 recordsLinked to original sources
arXiv2026

RelightFormer: Feed-forward Generative Transformer for Multiview Object Relighting

Image relighting is traditionally tackled via complex inverse rendering pipelines, which suffer from ill-posed optimization, or single-image generative models that ignore crucial multi-view cues necessary for understanding 3D geometry and material interactions. To address these limitations, we introduce a feed-forward generative Transformer for direct single- and multi-view image relighting that entirely bypasses explicit intrinsic property estimation. Adapted from a video foundation model, our architecture features a latent illumination module that dynamically injects target environment maps into spatial features via cross-attention. Furthermore, we employ permutation-invariant positional encodings to symmetrically process unordered multi-view inputs without sequential bias. To train this robust data-driven model, we construct the massive Laval Objaverse Dataset (LOD), comprising 90K objects and 39K unique illuminations. Extensive experiments demonstrate state-of-the-art visual quality, photorealistic relighting quality, and strong zero-shot generalization across single-view, multi-view, and novel-view relighting tasks.

cs.CV↗
arXiv2026

PhysInOne: Visual Physics Learning and Reasoning in One Suite

We present PhysInOne, a large-scale synthetic dataset addressing the critical scarcity of physically-grounded training data for AI systems. Unlike existing datasets limited to merely hundreds or thousands of examples, PhysInOne provides 2 million videos across 153,810 dynamic 3D scenes, covering 71 basic physical phenomena in mechanics, optics, fluid dynamics, and magnetism. Distinct from previous works, our scenes feature multiobject interactions against complex backgrounds, with comprehensive ground-truth annotations including 3D geometry, semantics, dynamic motion, physical properties, and text descriptions. We demonstrate PhysInOne's efficacy across four emerging applications: physics-aware video generation, long-/short-term future frame prediction, physical property estimation, and motion transfer. Experiments show that fine-tuning foundation models on PhysInOne significantly enhances physical plausibility, while also exposing critical gaps in modeling complex physical dynamics and estimating intrinsic properties. As the largest dataset of its kind, orders of magnitude beyond prior works, PhysInOne establishes a new benchmark for advancing physics-grounded world models in generation, simulation, and embodied AI.

cs.CV↗
arXiv2021

Asymptotic Convergence for a Class of Fully Nonlinear Contracting Curvature Flows

In this paper, we study a class of fully nonlinear contracting curvature flows of closed, uniformly convex hypersurfaces in the Euclidean space $\mathbb R^{n+1}$ with the normal speed $\Phi$ given by $r^\alpha F^\beta$ or $u^\alpha F^\beta$, where $F$ is a monotone, symmetric, inverse-concave, homogeneous of degree one function of the principal curvatures, $r$ is the distance from the hypersurface to the origin and $u$ is the support function of hypersurface. If $\alpha\geq \beta+1$ when $\Phi=r^\alpha F^\beta$ or $\alpha> \beta+1$ when $\Phi=u^\alpha F^\beta$, we prove that the flow exists for all times and converges to the origin. After proper rescaling, we prove that the normalized flow converges exponentially in the $C^\infty$ topology to a sphere centered at the origin. Furthermore, for special inverse concave curvature function $F=K^{\frac{s}{n}}F_1^{1-s}(s\in(0, 1])$, where $K$ is Gauss curvature and $F_1$ is inverse-concave, we obtain the asymptotic convergence for the flow with $\Phi=u^\alpha F^\beta$ when $\alpha=\beta+1$. If $\alpha<\beta+1$, a counterexample is given for the above convergence when speed equals to $r^\alpha F^\beta$.

math.DG↗
arXiv2021

Uniqueness and continuity of the solution to $L_p$ dual Minkowski problem

Lutwak, Yang and Zhang \cite{LYZ2018} introduced the $L_p$ dual curvature measure that unifies several other geometric measures in dual Brunn-Minkowski theory and Brunn- Minkowski theory. Motivated by works in \cite{LYZ2018}, we consider the uniqueness and continuity of the solution to the $L_p$ dual Minkowski problem. To extend the important work (Theorem \ref{uniquepolytope}) of LYZ to the case for general convex bodies, we establish some new Minkowski-type inequalities which are closely related to the optimization problem associated with the $L_p$ dual Minkowski problem. When $q< p$, the uniqueness of the solution to the $L_p$ dual Minkowski problem for general convex bodies is obtained. Moreover, we obtain the continuity of the solution to the $L_p$ dual Minkowski problem for convex bodies.

math.MG↗