Summary

As a PhD candidate at IGN—the French Mapping Agency 🗺—in the machine learning STRUDEL team (LASTIG lab, IGN-ENSG, Univ. Gustave Eiffel), and in the CSAI team at ENGIE Lab CRIGEN, I design deep learning methods for multimodal, multi-task learning on large-scale point clouds. Specifically, my recent research involves computer vision on 3D point clouds and 2D images in the wild.

You like point clouds ☁ ? You like images 📸 ? You like me 😊.



Supervisors



📃 Publications

🖼 Poster | 🎤 Oral

SuperCluster
Scalable 3D Panoptic Segmentation with Superpoint Graph Clustering
Damien Robert, Hugo Raguet, Loïc Landrieu
In Review

SuperCluster proposes a framework for efficient panoptic segmentation of large-scale point clouds. We formulate the panoptic task as a graph clustering problem. We train a model to predict desirable node and edge attributes to be used as input for a downstream graph clustering algorithm. This allows for training a model with only local supervision, without the need for non-maximum suppression, instance matching, and without any prerequisite on the number of objects in the scene. SuperCluster achieves new SOTA panoptic segmentation on indoor datasets S3DIS Area 5 (46.3 PQ (+4.0)) and ScanNetV2 (54.5 PQ (+21.0)), as well as outdoor datasets KITTI-360 (48.1 PQ) and DALES (53.9 PQ).
🦋 209k param. | ⚡ Train S3DIS F5 in 4h | 💾 20M-point inference on 1 GPU



superpoint transformer
Efficient 3D Semantic Segmentation with Superpoint Transformer
Damien Robert, Hugo Raguet, Loïc Landrieu
Paper | Webpage | Code
ICCV 2023

SPT is a superpoint-based transformer 🤖 architecture that efficiently ⚡ performs semantic segmentation on large-scale 3D scenes. This method includes a fast algorithm that partitions 🧩 point clouds into a hierarchical superpoint structure, as well as a self-attention mechanism to exploit the relationships between superpoints at multiple scales. We reach SOTA on S3DIS 6-Fold (76.0 mIoU), KITTI-360 Val (63.5 mIoU), and DALES (79.6 mIoU)n with:
🦋 212k param. | ⚡ Train S3DIS F5 in 3h | ⌚ SPG preprocessing ÷7



deepviewagg
Learning Multi-View Aggregation In the Wild for Large-Scale 3D Semantic Segmentation
Damien Robert, Bruno Vallet, Loïc Landrieu
Paper | Webpage | Code | Video
CVPR 2022 Oral 🎤 and Best Paper finalist 🎉

An end-to-end multi-view aggregation method for 3D semantic segmentation from images and point clouds. We reach SOTA on S3DIS and KITTI-360 without requiring point cloud colorization, meshing, or depth sensors: just point clouds ☁, images 📸, and their poses.



📑 Short Resume



🖼🎤 Talks & Presentations

🖼 Poster | 🎤 Oral



📚 Teaching

  • 01/2023 Deep Learning for Remote Sensing at ENSG (Course Instructor, M2, 13 hours)
  • 06/2022 3D Deep Learning for Remote Sensing at ISPRS 2022 (Tutorial Instructor, 1 day)
  • 05/2022 3D Deep Learning, Torch-Points3D & DeepViewAgg at ENGIE Lab CRIGEN (Tutorial Instructor, 1 day)
  • 01/2022 Deep Learning for Remote Sensing at ENSG (Course Instructor, M2, 9 hours)
  • 11/2020 Deep Learning for Computer Vision at Ecole Polytechnique (Teaching Assistant, M1, 12 hours)


🏠 Affiliations