Freelancer profile translated to English.

Description

Do you have images to analyze, objects to detect, or areas to segment? I build custom Computer Vision pipelines, from dataset to deliverable model.

Currently in an apprenticeship at Naldeo for almost 3 years, I have worked on real projects for local authorities and private clients: rooftop detection on satellite imagery (IoU 0.92), vegetation segmentation on 5 cm/pixel ortho-images, multi-object tracking in surgical video. Projects with real constraints, imperfect data, and expected deliverables.

I primarily work with PyTorch, YOLO, U-Net, and Transformer architectures (Swin). Comfortable with detection, instance/semantic segmentation, and tracking.

What sets me apart: I don't just deliver a model, I make sure it runs, the metrics are honest, and you understand what you have in your hands.

Available remotely for short or one-off missions.

Industry field of expertise

Languages

French
Native or bilingual
English
Fluent

Workplace preferences

Remote only

Primarily works remotely

Indépendant
Freelance Computer Vision Engineer
February 2026 - Today (6 months)
Lyon, France
Development of personal and experimental Computer Vision projects: detection and segmentation on satellite data, reimplementation of architectures from scratch, deployment of models via REST API. Projects available on GitHub.
Computer Vision Python Pytorch Deep Learning
Naldeo Digital for Climate,
Data Consultant
CONSULTING AND AUDITS
May 2024 - Today (2 years and 3 months)
Lyon, France
Computer vision applied to satellite and geospatial imagery for client projects. End-to-end pipeline: dataset construction, training, evaluation, delivery of outputs in GIS formats. Rooftank detection and water consumption fraud study (Amman, Jordan) 890 annotated images, comparison of YOLO vs RF-DETR, IoU 0.92. Cross-referenced detections with water consumption GIS layers to estimate median consumption per building and identify anomalies. Vegetated swale detection (Bordeaux Métropole, France) Segmentation on BD ORTHO ortho-images at 5 cm/pixel. Dataset of 1,210 images, comparison of Mask R-CNN vs U-Net, U-Net selected (IoU 0.76). Authored the final study delivered to the client to identify infrastructure ownership. Sewer infrastructure detection on Google Street View (PoC) YOLO pipeline applied to Street View imagery to automate field surveying. Web GIS application (QGIS Server) for local authorities; spatial database and interactive map for the Ardèche water supply master plan (SDAEP), with ETL scripts harmonizing 15 heterogeneous data sources.
Computer Vision QGIS Python Deep Learning Pytorch