Computer vision group of Astra Team, Inria Paris

Oct. 2023
We study 2D vision and 3D perception for robust scene understanding. Our research focuses on relaxing the use of abundant data and supervision, stepping towards weak-/un-supervised vision algorithms, while providing models that are more interpretable. We primarily address autonomous driving but our research expands to a variety of indoor and outdoor applications.

Research axes

Learning with less supervision
Learning with less supervision
Vision in complex conditions
Vision in complex conditions
3D scene understanding
3D scene understanding


Feb 26, 2024 Our 3 papers got accepted to CVPR 2024: 🎨 Material Palette, 🍴 FAMiX, 🕵️ PaSCo.
Dec 6, 2023 We just released three new works: 🎨 Material Palette for extraction of materials from a single image, 🍴 FAMiX a simple recipe for domain generalized segmentation, 🕵️ PaSCo the first 3D panoptic scene completion with uncertainty. Joint works with Oxford Uni.,, and TUM.
Dec 6, 2023 Opening of the internship seasons, check the job page. (18/12: 🔊 new internships added)
Oct 11, 2023 Our collaborative work 🧠 DREAM on visual decoding of brain activity is accepted in WACV 2024.
Oct 10, 2023 Intense week at ICCV23 with 2 papers (PØDA and SceneRF), 1 workshop keynote (R. de Charette @BRAVO), and 1 cool new group picture. Check this post for more info.
Sep 8, 2023 We organize the 2nd Weakly Supervised Computer Vision workshop at the 2023 Deep Learning Indaba in Accra, Ghana.
Jul 15, 2023 :loudspeaker: Two papers accepted for ICCV 2023: PØDA and SceneRF.
Jun 1, 2023 :raised_hands: Excited to welcome Tetiana Martyniuk soon starting a PhD in our :rocket: astra-vision team
May 1, 2023 Welcoming Tấn Khiêm Huỳnh and Souhaiel Ben Salem both joining for 6 month internships :tada:.
Nov 20, 2022 We opened 1 PhD and several internships More details
Oct 15, 2022 First group retreat, with intense brainstorming !
Aug 25, 2022 We organised a workshop on Weakly Supervised Computer Vision at Indaba Deep Learning
Jul 31, 2022 Astra-vision is created, the group has 5 Inria and 6 members