MrLabelling
Research platform for automatic 3D object segmentation and reconstruction from HoloLens 2 data. Combines YOLO, Mask R-CNN, and SAM achieving 80% precision for mixed reality applications.
Timeline
2024-12 — 2025-07
Technologies
Overview
3D Reconstruction for Mixed Reality
MrLabelling is a research project developed as part of my Master's thesis at UQAC. The platform automates 3D object segmentation and reconstruction from Microsoft HoloLens 2 sensor data, serving both smart environment MR applications and high-quality ML dataset creation.
80%
Precision
On custom HoloLens dataset
200+
Objects
Successfully reconstructed
3
Models
YOLO · Mask R-CNN · SAM
UCAml 2025
Publication
Springer LNNS
AI Pipeline
Multi-model segmentation
- •YOLO — Fast object detection, real-time performance, bounding box generation
- •Mask R-CNN — Pixel-perfect instance segmentation, multi-class detection
- •SAM (Segment Anything) — Zero-shot segmentation, interactive point-and-click annotation
Reconstruction Pipeline
From scan to 3D mesh
- 1Capture — HoloLens 2 spatial scanning with RGB-D sensor fusion
- 2Segmentation — AI model object detection and masking
- 3Annotation — Manual refinement if needed via the annotation platform
- 4Reconstruction — OccupancyNet deep learning mesh generation
- 5Validation — Quality assurance and precision metrics
- 6Export — OBJ, PLY, FBX formats
Academic Output
Published Research
Co-authored paper: Automatic 3D Object Segmentation and Reconstruction from HoloLens 2 Data for Mixed Reality in Smart Environments — submitted to UCAml 2025, 17th International Conference on Ubiquitous Computing and Ambient Intelligence. Published in Springer Lecture Notes in Networks and Systems.
More like this
Related work
ai
ClearSheet
Document scanning and OCR web app that detects a page inside a noisy phone photo, corrects perspective, generates a clean scan-style result, and extracts text in French, English, or both.
ai
ThoraxVision
AI-powered web application for detecting pulmonary diseases from chest X-rays using deep learning. YOLO-based detection with 80% accuracy on the NIH dataset.