Mikhail Derevyannykh

I'm Mikhail Dereviannykh or just Mike

I do a PhD in Neural Rendering at KIT. My primary research interests are: efficient neural network inference, world models, and e2e rendering algorithms.

I was lucky to be mentored and supervised by Tobias Ritschel (UCL), Anton Kaplanyan (Meta, Nvidia, Intel), Dario Seyb (Valve, Meta), Carsten Dachsbacher (KIT), Johannes Hanika(KIT, Weta Digital), Mark Boss (Unity, Stability AI), Shimon Vainer (Unity, Stability AI), and many others!

Previously, I worked in gamedev, focusing on graphics, rendering, and AI

from 05.2026 Intern Research Scientist, World Models, Stability AI
07.2025 - 05.2026 Research Scientist Intern, Meta Reality Labs, Redmond, USA
2022–Present – PhD Researcher, KIT (Prof. Dachsbacher), Karlsruhe, Germany
2021–2022 – Researcher, MSU Graphics & Media Lab, Moscow
2018–2022 – Senior R&D Engineer, Eagle Dynamics (DCS World), Moscow
2015–2017 – Founder & CTO, WellDone Team, Moscow

devmishok@gmail.com linkedin twitter bsky

Updates:
- Invited to present "Neural Two-Level Monte Carlo Real-Time Rendering" at SIGGRAPH 2025 [New] [Link] - August 2025
- Secured Research Scientist Internship at Meta, Reality Lab, under Dr. Dario Seyb's supervision - Fall 2025
- Received "Best Paper Honorable Mention" at Eurographics 2025 [New] [Link] - 16.05.25 ⭐
- Meshy Fellowship 2025 Outstanding Prize Recipient [New] [Link] - 20.02.25 💸
- I started a PhD at KIT, IVD [Link] - 22.07.22
- Gave a talk about "Real-Time Neural Graphics" at Intel, GPU Research - 05.01.22

On Cosine Prior Distributions for Neural Path Guiding

Jan-Luca Gutsch, Mikhail Dereviannykh, Johannes Hanika
Eurographics 2026

TLTR: We show that Light-Transport-inspired priors help Normalizing Flows to train path-guiding distributions faster and more accurately. It’s basically the anti-Bitter Lesson for tiny models

The work was primarily done by my student Jan-Luca Gutsch. I just led it.

Paper

Neural Two-Level Monte Carlo Real-Time Rendering

Mikhail Dereviannykh, Dmitrii Klepikov, Johannes Hanika, Carsten Dachsbacher
Eurographics 2025, Computer Graphics Forum ⭐Best Paper Honorable Mention⭐ SIGGRAPH 2025

TLTR: We introduce Neural Incident Radiance Cache (NIRC) enabling efficient integration over a neural network domain instead of relying on costly path tracing. To ensure unbiased estimations and improved accuracy, we employ a Two-Level Monte Carlo approach, outperforming Neural Control Variates in training speed and representative power, saving on RTX usage

Paper | Project Page

Real-Time Path-Guiding Based on Parametric Mixture Model

Mikhail Derevyannykh
Eurographics 2022

TLTR: The first lightweight screen-space path-guiding for real-time use, based on online learning of parametric mixture models. 4× FLIP metric reduction at 1 spp, runs in <1.5ms on RTX 2070 at 1080p, and improves path coherence by 5%, reducing bias and SVGF flickering

Conference Version | Fast Forward | Video Presentation | Video Demo

DCS World VFX System - Senior Graphics Programmer

Developed a real-time GI algorithm (0.1ms update) using neural networks and distributed computing
Researched path-guiding for real-time rendering, presented at Eurographics 2022.
Built and optimized graphics effects, PBR volumetrics, and a particle editor.
Boosted engine performance with NSight, VTune, and C++/LUA optimizations.

Video Steam

Stochastic Spherical Gaussians Radiance Volumes For

Production Real-time Rendering

We present the results of a study of the fast, flexible, and robust solution of the global illumination problem based on precomputed light volumes with stochastically filtered Spherical Gaussians (SG). It supports coherent shading of dynamic and static objects, taking into account visibility weighting for preventing light and shadow leaks. In addition, our lighting basis and optimization algorithm of its parameters are robust for high-frequency indirect lighting and achieve lower errors concerning the current state-of-the-art methods based on SG. Furthermore, we suggest a more accurate approach for computing Fresnel, Smith Masking Function, and Cos factors of light transport equation when incident lighting is approximated by SG basis. And finally, a method for estimating local high-frequency occlusion with SG basis is presented, too.

Paper (Small) Paper Video[Sponza] Video[Cockpit]

Hungry Jump - CTO/Co-Founder of WellDone Games

Led development, ensuring smooth shipping, QA, and game design integration.
Delivered a polished game with 100 levels, 16 unique AI-driven insects, and 10+ custom UI screens.
Built a scalable data system (Google Spreadsheets + Python) managing 1M+ logic cells.
Engineered advanced insect flight physics using Markov Chain Random Walk.
Optimized performance and animations with Cocos2d-x + GAFMedia.

Video Google Play