Research Radar

arXiv cs.GR details a diffusion-stylization method that upgrades two existing conditioning weights from global sliders to per-region spatial maps. The work is aimed at the practical pain point of professional retouching: current latent-diffusion stylization pipelines tend to entangle what a region shows with how it …

• A diffusion stylization method adds local, per-region control over content and style using spatial maps.
• Amir Semmo published the work on arXiv cs.GR.

arXiv cs.GR details PrimitiveCAD, a new point-to-CAD reconstruction method aimed at making LLM-driven CAD generation more faithful to real geometry. The work targets a common weakness in current approaches: treating CAD reconstruction like generic point-cloud token prediction instead of modeling CAD primitives and operations directly. …

• PrimitiveCAD is a multi-stage point-to-CAD reconstruction method using primitive-aware tokenization, LLM fine-tuning, and RL.
• Authors: Jian Gao, Kailin Bi, Jiamin Xu, Jinlan Xu, and Gang Xu.

arXiv cs.GR details UltraDiff, a differentiable ray-tracing framework that extends physically based optimization from light transport to medical ultrasound. The core idea is to model image formation as a path-space integral gated by travel time, then derive gradients that let the system tune scene parameters …

• UltraDiff is a differentiable ultrasound ray-tracing framework for shape optimization.
• Authors: Felix Duelmer, Magdalena Wysocki, Nassir Navab, and Mohammad Farid Azampour.

arXiv cs.GR details Neuroll, a neural strand-based hair simulation method built around simulator-in-the-loop unrolling. The core idea is to replace expensive classical time integration with a learned integrator that takes prior hair state, material stiffness, and collision geometry as inputs. For developers, the interesting part …

• Neuroll is a real-time neural strand-based hair simulation method using simulator-in-the-loop unrolling.
• Authors: Gene Wei-Chin Lin, Jessica Jia-En Lee, Yu Ju (Edwin) Chen, Egor Larionov, and Tuur Stuyck.

arXiv cs.GR details PhysLDM, a new latent-diffusion approach for one-shot volumetric deformable simulation aimed at the hard problem of predicting high-resolution 3D mesh motion over long horizons. The pitch is straightforward: instead of simulating every frame at native resolution, the system learns a compact spatiotemporal …

• PhysLDM is a latent-diffusion model for high-fidelity deformable simulation
• Yu Zhang, Xudong Xu, and Xingang Pan

arXiv cs.GR details View Matters, a text-driven 3D Gaussian editing framework that tries to solve a familiar problem: not every camera view is equally useful for guiding an edit. The method focuses on reliable keyframes rather than averaging supervision across all rendered views, which is …

• View Matters is a keyframe-guided, text-driven 3D Gaussian editing framework
• Kaizhe Zhang, Yijie Zhou, Weizhan Zhang, Xuanyu Wang, Feng Lei, and Sha Gong

arXiv cs.GR details Humanoid Horizon, a new policy framework for humanoid robots that tackles long-horizon loco-manipulation in cluttered indoor spaces. The goal is to let a single agent navigate, grasp, carry, and place multiple heavy objects in sequence without earlier progress being undone as the …

• Humanoid Horizon is a unified policy framework for long-horizon humanoid loco-manipulation.
• Paper by Haozhuo Zhang and six coauthors, published on arXiv cs.GR.

arXiv cs.GR details PDB, or Point-Based Deformation Blending, a new facial animation retargeting approach built to move expressions between meshes with different topology. The key idea is to predict a compact set of deformed control points from a source neutral-expression pair, then reuse target-specific blending …

• PDB is a point-based deformation blending method for facial animation retargeting across different meshes.
• Authors: Sihun Cha, Hyeonseung Shin, Suah Yu, and Junyong Noh.

arXiv cs.GR details 4D-HOF, a feed-forward framework for 4D hand-object interaction reconstruction that starts from coarse outputs from vision foundation models and refines them toward a learned interaction manifold. The paper is aimed at a long-standing pain point in capture and reconstruction work: getting stable …

• 4D-HOF is a feed-forward 4D hand-object interaction reconstruction method
• Authors: Shiqi Li, Sean Cho, Yijie Li, Fengzhi Guo, Bowen Wen, Cheng Zhang

arXiv cs.GR details SteadySplats, a new approach to stochastic rendering for 3D Gaussian Splatting that aims to make order-independent transparency practical without the usual speckled output. The paper frames visible noise as the main blocker and attacks it at both the representation and image-synthesis levels. …

• SteadySplats proposes low-variance resampling for high-fidelity stochastic rendering of 3D Gaussian Splatting.
• Authors include Felix Windisch, Thomas Köhler, Lukas Radl, Chris Wyman, Georgios Kopanas, Bernhard Kerbl, and Markus Steinberger.

arXiv cs.GR details a new geometric measure for speech-driven 3D facial animation that focuses on coarticulation, the subtle motion shaping that happens as sounds influence one another. The idea is to compare lip-path length against a consonant-aware route through vowel, consonant, and vowel poses, rather …

• A geometric metric for coarticulation in speech-driven 3D facial animation
• Danzel Serrano and Przemyslaw Musialski

arXiv cs.GR introduces EditHero, a benchmark aimed at a problem game teams actually face: assets rarely get edited once, they get revised over and over. The benchmark focuses on long-horizon, part-level 3D editing, with natural-language instructions and target images covering both geometry and texture. A …

• EditHero is a new benchmark for long-horizon, part-level 3D editing and vibe modeling.
• Paper by Ruihan Yu and 11 coauthors, published on arXiv cs.GR.