Site logo: PCB with AI brain

Computer Vision Lab

Nikolay Falaleev

Profile

Nikolay S. Falaleev

London, United Kingdom, nikolasent@gmail.com

Founder of Fanis AI, SVP of AI & Computer Vision at Sportlight, Kaggle Master

fanis.ai · GitHub · LinkedIn · Google Scholar

Visionary Computer Vision researcher and founder of Fanis AI, an independent lab that takes state-of-the-art methods from research into production for problems where accuracy is critical. Over ten years, built and deployed real-time vision systems ranging from multi-camera and LiDAR player tracking in every English Premier League stadium to scientific image denoising. Recent research published at CVPR and ICCV.

Skills & Expertise

Computer Vision:   Object Detection & Tracking, Semantic Segmentation, Depth Estimation, Pose Estimation, Video Event Detection, 3D Scene Reconstruction, Image Restoration & Denoising, Multi-Modal Sensor Fusion, Localization & Path Planning
Imaging & Signal Processing:   Camera Calibration, Image Signal Processing (ISP), Debayering, Image Enhancement, Kalman Filtering, Fourier Analysis, Numerical Methods, Statistical Modeling, Optimization
Real-Time & Edge AI:   Real-time video pipelines optimised for both accuracy and speed, on-device inference, TensorRT, ONNX, TFLite, OpenVINO, DeepStream, GStreamer
AI/ML Technologies:   PyTorch, TensorFlow, Scikit-learn, NumPy, OpenCV, FFmpeg
Hardware:   Industrial Cameras, Optics, LiDARs, Depth Cameras, IMUs, DL Accelerators
MLOps & Deployment:   MLflow, gRPC, FastAPI, Docker, Cloud Infrastructure (Azure, AWS, DigitalOcean), CI/CD Pipelines
Multi-Modal Agentic AI:   vLLM, LangChain, vector DBs, AI agents over enterprise data
Languages:   Python, C/C++, R, SQL
Leadership:   AI Strategy & R&D Direction, Technical Leadership, Cross-Functional Team Leadership, Scaling AI Solutions, Mentoring

Experience

02/2025 - present, Founder & Head of AI Fanis AI

Independent R&D lab for precision Computer Vision and AI, from research to production

  • Delivering Computer Vision projects for clients across the globe where accuracy is critical, taking state-of-the-art results from research to production.
  • Developed FastSHADE, a real-time image denoising model for mobile GPUs that set a new SOTA in image quality (PSNR) on the MAI2021 benchmark; Top-2 in the Mobile AI 2026 Real Image Denoising Challenge (CVPR 2026).
  • Research on scientific imaging: the first application of diffusion models to restoring electron backscatter diffraction (EBSD) patterns, in association with AMOLF (Netherlands), published at ICCV 2025.
  • Multi-modal agentic AI for enterprise, e.g. natural-language chat over enterprise SQL databases.

Working on a Computer Vision or AI problem where accuracy matters? Get in touch: hello@fanis.ai · fanis.ai

10/2020 - present, SVP of AI & Computer Vision (07/2026 - present), Head of AI (10/2020 - 06/2026) Sportlight Technology

AI-Driven Analytics for Elite Sports (EPL, SPL, NBA, MLB, NHL, and others)

  • Lead technology across hardware, software and AI teams, taking projects from idea to commercial operation: hardware design, scalable backend, and AI model design and architecture for gold-standard accuracy of player tracking and sports analytics data.
  • Defined the company’s AI strategy and led the development of a real-time system that fuses multi-camera video and LiDAR data (object detection, tracking, pose estimation, action spotting).
  • Architected and deployed an operational AI system used across all English Premier League (EPL) stadiums, the Saudi Pro League (SPL), and other venues worldwide, achieving industry-leading player tracking accuracy and reducing data QA time and costs by 12x.
  • Led the transition to real-time data processing, enabling deployments across various sports and geographies and scaling AI operations by 50x.

08/2017 - 08/2020, Leading Computer Vision R&D Constanta/OSAI

Led the research, development, and deployment of cutting-edge Computer Vision systems for real-time sports analytics, transforming data capture and analysis for tennis, basketball, billiards, and esports. Spearheaded projects that replaced human scouting with automated, highly accurate CV solutions, driving significant cost savings and enhancing data quality for betting, coaching, and broadcast applications.

  • Pioneered a real-time spatio-temporal video processing architecture, published in a CVPR 2020 paper. The system was showcased at high-profile events: Tokyo 2020 Olympics, JOOLA North American Teams Championships and National Table Tennis Championship 2020, and has been deployed in numerous venues worldwide for 24/7 commercial operation. The system is 52% faster than humans, 40% cheaper and has accuracy above 97.5%. SBJ article.
  • Architected and deployed Computer Vision modules for real-time video analysis of 100,000s of eSports tournaments monthly, including FIFA, NHL, and NBA.
  • Led the end-to-end development of a Computer Vision module for pool-9 real-time analysis, overseeing the entire project from data collection to product deployment. This system delivered significant cost savings. Video demo, Blog article [rus].

10/2014 - 07/2017 Junior Researcher Academic Research Institute

Developed and analyzed stable finite-difference schemes for modeling transient heat and mass transfer in three-dimensional geometries using C/C++. Contributed to six peer-reviewed scientific publications and presented findings at three conferences.

Selected Publications

  1. Falaleev N., FastSHADE: Fast Self-augmented Hierarchical Asymmetric Denoising for Efficient inference on mobile devices, // The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops Proceedings, 2026, ArXiv:2604.10275.

  2. Falaleev N., Orlov N. DIFFRACT: Diffusion-based Restoration via Adaptive Control and Thresholding for Diffraction Imaging, // Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2025, pp. 5579-5588, DOI: 10.1109/ICCVW69036.2025.00584, ArXiv:2504.16951.

  3. Falaleev N., Chen R. Enhancing Soccer Camera Calibration Through Keypoint Exploitation, // MMSports ‘24: Proceedings of the 7th ACM International Workshop on Multimedia Content Analysis in Sports, 2024, pp. 65-73, DOI: 10.1145/3689061.3689074, GitHub repo.

  4. Cioppa A., Giancola S., …, Falaleev N., et al. SoccerNet 2023 challenges results, // Sports Engineering, V. 27, 2024, DOI: 10.1007/s12283-024-00466-4, ArXiv:2309.06006.

  5. Voeikov R., Falaleev N., Baikulov R. TTNet: Real-time temporal and spatial video analysis of table tennis, // The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops Proceedings, 2020, pp. 884-885, DOI: 10.1109/CVPRW50498.2020.00450, ArXiv:2004.09927.

Google Scholar profile

ResearchGate profile

Patents

  • RU106765 “Device for producing 3D images by a single camera”. A method for calculating depth maps from images produced by the device was also developed, and the device can also work as a range finder.

Public Talks

06/2026     CVPR 2026 Workshops
      FastSHADE: real-time image denoising for efficient inference on mobile devices. Presentation of the paper ArXiv:2604.10275.
11/2024     London PyTorch Meetup
      Optimising Video Pipelines for Neural Network Training with PyTorch. Slides, GitHub repo.
11/2024     Computer Vision Summit London 2024
      Transforming Athlete Performance with LiDAR and Video Data. Keynote presentation, 250+ attendees.
11/2023     Computer Vision Summit London 2023
      Deploying unparalleled accuracy in athletes’ performance analysis with LiDAR and video data processing. Keynote presentation, 250+ attendees.
06/2023     CVPR, 9th International Workshop on Computer Vision in Sports
      Presentation on a DL approach to camera calibration. GitHub repo.
07/2022     GSIG Solutions Showcase: Athlete tracking: AI, computer vision, machine learning
      LiDAR and visual data fusion for athlete tracking.

Prizes and Awards

2026     Mobile AI 2026 Real Image Denoising Challenge, CVPR
      Top-2, with the best PSNR score. FastSHADE model for real-time image denoising on mobile GPUs. ArXiv:2604.10275.
2023     SoccerNet Camera Calibration Challenge 2023, CVPR
      Top-1. DL approach to camera calibration from football broadcast videos. The challenge was held at CVPR 2023.
2018     Quick, Draw! Doodle Recognition Challenge, Kaggle
      Top-4% (50/1316). Application of deep neural networks to the classification of users’ drawings into 300+ classes with 50M training samples.
2018     TGS Salt Identification Challenge, Kaggle
      Top-0.5% (14/3234), Gold Medal. A deep-learning approach to segmenting salt deposits in seismic images, which is crucial for oil and gas drilling.
2018     Lyft Perception Challenge, Udacity
      Top-3% (4/155) and the fastest neural network pipeline. The solution is based on LinkNet34 for real-time multiclass semantic segmentation.
2018     IEEE’s Signal Processing Society - Camera Model Identification, Kaggle
      Top-3% (15/583). The solution uses convolutional neural networks to identify the camera model that captured a given image.
2016     Image-Based Localization Challenge, Udacity
      3rd place. A deep learning approach based on the GoogLeNet Inception module, using a “localization as classification” method to localize an autonomous vehicle.

Education

09/2015 - 07/2017     Master of Science with Honours in Materials Science, Department of Materials Science
      Lomonosov Moscow State University (MSU), Grade: 5.0/5.0
09/2011 - 07/2015     Bachelor of Science in Materials Science, Department of Materials Science
      Lomonosov Moscow State University (MSU)

Continuing education

07/2021 - 08/2021 Oxford Machine Learning Summer School 2021
  Certificate
03/2021 - 08/2021 Intel Edge AI for IoT Developers Nanodegree
  Udacity, Certificate
01/2020 - 06/2020 C++ Nanodegree
  Udacity, Certificate
11/2016 - 10/2017 Self-Driving Car Engineer Nanodegree
  Udacity, Certificate, graduated with the first ever cohort
11/2012 - present >30 MOOCs in Computer and Data Sciences, Robotics, Computer Vision and Machine Learning
  Coursera, Udacity, edX, etc.

Clubs & Societies

Founded a non-commercial organization OniroAI (Github) for independent research and development in Computer Vision and Artificial Intelligence.

Other Publications

Nanotechnology:

  1. Eliseev Andrei A., Kumskov A.S., Falaleev N.S., Zhigalina V.S., Eliseev Artem A., Mitrofanov A.A., Petukhov D.I., Vasiliev A.L., Kiselev N.A. Mass Transport Through Defects in Graphene Layers // Journal of Physical Chemistry C, 2017, DOI: 10.1021/acs.jpcc.7b06100

  2. N. S. Falaleev, A.S. Kumskov, V.G. Zhigalina, I.I. Verbitskiy, A.L. Vasiliev, A. A. Makarova, D. V. Vyalikh, N.A. Kiselev, A.A. Eliseev. Capsulate structure effect on SWNTs doping in RbxAg1−xI@SWNT composites // CrystEngComm, 2017, DOI: 10.1039/c7ce00155j

  3. Andrei A. Eliseev, Nikolay S. Falaleev, Nikolay I. Verbitskiy, Andrei A. Volykhov, Lada V. Yashina, Andrei S. Kumskov, Victoria G. Zhigalina, Alexander L. Vasiliev, Alexey V. Lukashin, Jeremy Sloan, Nikolay A. Kiselev. Size-dependent structure relations between nanotube and encapsulated nanocrystal. // Nano Letters, V. 17 I. 2, 2017, P. 805–810 DOI: 10.1021/acs.nanolett.6b04031

  4. N.A. Kiselev, A.S. Kumskov, V.G. Zhigalina, A.L. Vasiliev, J. Sloan, N.S. Falaleev, N.I. Verbitskiy, A.A. Eliseev. The structure and continuous stoichiometry change of 1D TbBrx@SWCNTs // Journal of Microscopy. V. 262, I. 1, April 2016, P. 92–101 DOI: 10.1111/jmi.12348

  5. Lukashin A.V., Falaleev N.S., Verbitskiy N.I., Volykhov A.A., Verbitskiy I.I., Yashina L.V., Kumskov A.S., Kiselev N.A., Eliseev A.A. Quasi free-standing one-dimensional nanocrystals of PbTe grown in 1.4 nm SWNTs // Nanosystems: physics, chemistry, mathematics. V. 6, I. 6, 2015, P. 850-856 DOI: 10.17586/2220-8054-2015-6-6-850-856