Keywords: Computer Vision, Photogrammetry, Machine Learning, Place Recognition, Pose Estimation, 3D Points Clouds, Big Data, LLM

Full description and candidature: https://www.umr-lastig.fr/vgouet/News/sujet_stage_2025-Loc3D-ext-v2.pdf 

Place recognition based on the visual mapping of the environment is a problem at the heart of many topical application domains, such as geolocalization for mobile mapping, digital twins update and documentation, collections annotation in digital humanities, augmented reality or fact-checking. Recognizing a location can take many forms, from the production of an annotation to a 6D pose that also provides information on the location of the acquisition sensor. In the state of the art of computer vision, when no initial position is known, existing techniques are based on indexing and similarity search of visual content in a geolocalized image repository. Here, we study the generalization of this type of approach to 3D by considering 3D point cloud acquisition campaigns (notably LiDAR), which are becoming increasingly popular and whose richness in terms of geometry and semantics is attractive, but with a volume and diversity that are complex to handle. The internship is at the heart of the problem of indexing and retrieval in 3D point clouds for place recognition, through the study of deep 3D points cloud descriptors up to efficient retrieval and reranking for 3D pose estimation.

Bac+5 in computer science, applied math or computer vision (master or engineering school); good knowledge in image or 3D data processing, as well as strong skills in Python programming. Good skills in Apache Spark, hugging Face API, LLM, PyTorch, or functional programming is a significant plus.

Before February 15th 2025, send by e-mail to the contacts in a single PDF file: