Nicolas ADAMCZYK
Computer Vision Student at Université Paris Cité
Based in Paris, France
Leveraging C++, Python, and Unreal Engine 5 to bridge the gap between physically-accurate rendering and robust AI models.
Self-taught 3D artist since 2023. I leverage my expertise in lighting and realistic rendering to build physically-accurate virtual environments. My focus is training robust AI models using perfectly controlled synthetic datasets.
Experience
CS Studying / Since 2020
▹ 3D Modeling / Since 2022
▹ CV & AI / Since 2025
Communication
FR / EN / PL
Core Stack
Python, C++, Blender, UE5
▹ PyTorch, OpenCV, CUDA
Most
Recent Projects
Last Update05/03/2026
Euro Coin Detection & Identification
Euro Coin Detection & Identification
- ▹Hybrid detection system combining Hough Circle Transform, Watershed segmentation, and Contour analysis.
- ▹Robust preprocessing with CLAHE and HSV/Lab color space analysis for adaptive background handling.
- ▹Classification via KNN and rule-based logic to distinguish between Copper, Gold, and Bimetallic groups.
- ▹Optimized performance reducing processing time from 60s to 0.24s per image.
OpenCVPythonNumPyScikit-Learn
Orbital ISS 3D Tracker
Orbital ISS 3D Tracker
- ▹High-fidelity 3D Earth rendering using Three.js with custom shaders for day/night cycles.
- ▹Real-time orbital calculation using SGP4 propagation models and TLE data fetch.
- ▹Interactive UI for live telemetry: altitude, velocity, and geographic coordinates.
- ▹Responsive orbital path prediction with dynamic camera transitions and Earth-locked views.
Three.jsReactSGP4WebGL
Check full repository on Github
Activity
Currently Building
Last Update04/03/2026
SATELLITA : Orbital Network Simulator
SATELLITA : Orbital Network Simulator
Development Phase
Feb 2026 - July 2026 (Est.)Independent / Personal Lab
- ▹Architecting a high-performance aerospace simulator using Unreal Engine 5.4 and Cesium for Unreal plugin.
- ▹Implementing a scalable C++ backend for real-time SGP4 orbital propagation and UTC-synchronized Time Management.
- ▹Developing an advanced 'Pawn' navigation system for seamless transition between global Earth views and localized satellite tracking.
- ▹Designing a modular subsystem architecture to handle large-scale satellite constellations and space debris datasets.
Research : ShapeEmbed & Contour Quantification
Research : ShapeEmbed & Contour Quantification
Development Phase
Jan 2026 - Mai 2026 (Est.)Université Paris Cité
- ▹Developing a self-supervised representation learning framework based on the ShapeEmbed architecture for 2D contour encoding.
- ▹Implementing Euclidean Distance Matrices (EDM) to ensure shape descriptors are invariant to translation, rotation, and scaling.
- ▹Training a Variational Auto-Encoder (VAE) to map complex biological and natural shapes into a compact latent space.
- ▹Optimizing the reconstruction path to generate synthetic data outlines for AI model training augmentation.