Colloquia and Seminars

Room impulse responses provide an indirect acoustic probe of scene geometry: as an agent moves, the recorded reverberation changes with nearby walls, openings, and free space. However, converting acoustic observations into maps is inherently ambiguous, since different reflector configurations can produce similar responses, especially when observations are sparse or motion is limited. 

We study active acoustic scene reconstruction, where an agent must choose sensing poses that improve its geometric belief rather than follow a predefined scan. We introduce a cross-modal acoustic world model that encodes histories of RIRs and known poses into a motion-conditioned latent state used for both local occupancy decoding and future acoustic-latent prediction. At test time, candidate trajectories are rolled out in latent space, decoded into imagined occupancy maps, and scored by predicted map-space information gain. We construct a synthetic benchmark of paired acoustic trajectories, poses, and floor-plan geometry. Experiments show improved local acoustic-to-geometry reconstruction over geometric and passive baselines, and closed-loop mapping that matches or improves frontier-based exploration while substantially reducing collisions.

Accurate real-time rendering under complex illumination remains a central challenge in computer graphics, particularly for interactive applications such as gaming. While image-based lighting (IBL) frameworks and the widely used SplitSum approximation enable efficient rendering of specular materials, they rely on simplifying assumptions that introduce artifacts, especially at grazing angles. In this work, we present a novel rendering scheme that improves the approximation of specular reflectance by leveraging spherical harmonics (SH) for low-frequency illumination. For the high-frequency components, we retain the efficiency of the SplitSum method. Our approach decomposes Env. Maps into frequency bands and introduces an analytic SH-based solution to the rendering equation in a unique local shading frame (ULSF), allowing BRDF representations to be efficiently stored in scene-independent lookup tables.We implement our method in Unity and OpenGL, demonstrating improved accuracy over prior approaches, including reduced light leakage and more faithful rendering of rough materials, while maintaining real-time performance with only marginal overhead compared to SplitSum

Yam Kushinsky is a PhD student at the Weizmann Institute of Science, advised by Prof. Ronen Basri. His research focuses on inverse rendering, Gaussian splatting, and real-time rendering — with a particular interest in recovering geometry, materials, and lighting from images.

Prior to his PhD, Yam worked for five years as an algorithms developer at Common Ground, where he focused on inverse rendering for human avatars. He holds a Bachelor’s degree in Physics and Electrical Engineering, and a Master’s degree in Applied Mathematics from the Weizmann Institute, where he studied convex optimization under the supervision of Prof. Yaron Lipman.

Selected for an oral presentation at CVPR 2026; Image classification is a well-studied task in computer vision, and yet it remains challenging under high-uncertainty conditions, such as when input images are corrupted or training data are limited. Conventional classification approaches typically train models to directly predict class labels from input images, but this might lead to suboptimal performance in such scenarios. To address this issue, we propose Discrete Diffusion Classification Modeling (DiDiCM), a novel framework that leverages a diffusion-based procedure to model the posterior distribution of class labels conditioned on the input image. DiDiCM supports diffusion-based predictions either on class probabilities or on discrete class labels, providing flexibility in computation and memory trade-offs. We conduct a comprehensive empirical study demonstrating the superior performance of DiDiCM over standard classifiers, showing that a few diffusion iterations achieve higher classification accuracy on the ImageNet dataset compared to baselines, with accuracy gains increasing as the task becomes more challenging.