I am a co-creator and long-term maintainer of Albumentations, and the founder of AlbumentationsX. I work on the libraries, benchmarks, documentation, and public evidence that help engineers and researchers build reliable computer-vision training pipelines.
I hold a PhD in physics and spent many years in industry as a data scientist and machine-learning engineer, building production computer-vision systems. I became a Kaggle Grandmaster through competitive machine learning; my earlier open-source work includes TernausNet and competition solutions for medical and satellite imagery.
My publications span theoretical physics, computer vision, autonomous driving, medical imaging, and satellite imagery. Google Scholar reported more than 8,800 citations, an h-index of 19, and 21 works cited at least ten times when checked on 22 July 2026.
| 8,856 citations |
19 h-index |
21 works with 10+ citations |
| Selected publication | Research area | Google Scholar citations on 22 July 2026 |
|---|---|---|
| Albumentations: Fast and Flexible Image Augmentations | Open-source computer vision | 3,734 |
| TernausNet: U-Net with VGG11 Encoder Pre-Trained on ImageNet for Image Segmentation | Image segmentation | 1,035 |
| One Thousand and One Hours: Self-driving Motion Prediction Dataset | Autonomous driving | 672 |
| Automatic Instrument Segmentation in Robot-Assisted Surgery Using Deep Learning | Medical imaging | 553 |
| Superconducting Transitions in Flat-Band Systems | Theoretical physics | 148 |
| Project | What I work on |
|---|---|
| AlbumentationsX | The actively developed augmentation library: releases, performance, annotation safety, reviews, and maintenance. |
| Albumentations documentation | Practical guides, transform reference, examples, and reproducible benchmarks. |
| Explore | A browser playground for testing transforms on your own images before adding them to a pipeline. |
| Albumentations adoption evidence | Verifiable examples of how open-source projects, researchers, and competition teams use the ecosystem. |
| 160.7M PyPI downloads |
15.3k GitHub stars |
40k+ public dependents |
2,270 citing research works |
These numbers describe public ecosystem adoption, not paid customers or endorsements. They were checked against PyPI, GitHub, and the deduplicated Albumentations adoption snapshot on 22 July 2026. Albumentations is a NumFOCUS Affiliated Project.
Open-source maintenance is recurring work: reviewing issues and pull requests, shipping releases, testing compatibility, rerunning benchmarks, improving documentation, and helping users diagnose difficult pipelines.
If Albumentations, TernausNet, or research I contributed to has helped your work, sponsorship helps fund the time I spend maintaining and extending open-source tools for the community.
| Project | Why it is here |
|---|---|
| TernausNet | A U-Net-style segmentation model with a VGG11 encoder, released with pretrained weights. |
| Robotic instrument segmentation | The winning solution and follow-up work for the MICCAI 2017 robotic instrument segmentation sub-challenge. |
| TernausNetV2 | A fully convolutional network for instance segmentation. |
| DSTL satellite imagery | Code from a third-place solution among 419 teams in the Kaggle challenge. |
- Open-source and project correspondence: [email protected]
- Research profile: Google Scholar
- Professional profile: LinkedIn
- Competition work: Kaggle
- Short updates: X / Twitter