PixSDS: Why Latent SDS Makes Noisy Pixels
ECCV 2026 LoViF Workshop
project page / paper
Identify the real reason behind structured noise in SDS generations.
**My first sole-authored paper!**
Vsevolod Skorokhodov
I am an M.Sc. student in Data Science at EPFL, with a strong interest in Computer Vision, 3D, Generative Modeling, and Efficient Inference.
Recently, I interned at the Schindler EPFL Lab, where I worked on multimodal foundation geometry models and on developing methods to generate simulation-ready 3D models directly from RGB images. Prior to this, I was a research student at CVLab under the supervision of Prof. Pascal Fua, where I focused on Gaussian splatting and diffusion models. Before joining EPFL, I earned my B.Sc. in Applied Mathematics and Physics at MIPT. During my undergraduate studies, my research centered on Uncertainty Estimation in Semantic Neural Radiance Fields.
I open to new connections and collaborations. Feel free to reach out via email or connect with me on GitHub or Telegram.
Projects
Gator: Cross-view Jigsaw Objective for Self-Supervised 3D Vision
Use Cross-View Jigsaw Puzzle Solving as a Self-Supervised Objective for 3D Vision.
SEAR: Simple and Efficient Adaptation of Visual Geometric Transformers for Unpaired RGB+Thermal 3D Reconstruction
ECCV 2026
Estimates 3D geometry of unpaired RGB and thermal frames.
D3DR: Lighting-Aware Object Insertion in Gaussian Splatting
TMLR 2026
Lighting correction makes inserted 3D Gaussian Splatting objects blend naturally into their new scenes.
uSF: Learning Neural Semantic Field with Uncertainty
Optical Memory and Neural Networks, 2024
A neural semantic field that estimates uncertainty alongside color and semantic labels.