Our paper "LiDPM: Rethinking Point Diffusion for Lidar Scene Completion" got accepted to IEEE IV 2025!

tldr: LiDPM enables high-quality LiDAR completion by applying a vanilla DDPM with tailored initialization, avoiding local diffusion approximations.

Project page: astra-vision.github.io/LiDPM/

Presenting our project #LiDPM in the afternoon oral session at #IV2025!

Project page: astra-vision.github.io/LiDPM/

w/ @gillespuy.bsky.social, @alexandreboulch.bsky.social, Renaud Marlet, Raoul de Charette

Also, see our poster at 3pm in the Caravaggio room and AMA 😉

We just released the code of #LiDPM, go ahead and play with it (and don't forget to star 🤭🤩)!

Training and inference code available, along with the model checkpoint.

Github repo: github.com/astra-vision...

#IV2025

Presenting LiDPM today at #CVPRinParis, such a great poster session, many new acquaintances and curious questions! #cvpr #cvpr2025

Next stop: #IV2025.

Hey #IV2025, @t-martyniuk.bsky.social is presenting her LiDPM work today!
tl;dr: LiDPM - a point diffusion strategy for scene completion from outdoor LiDAR point clouds.
Check out the paper and code below if you can't make it for the poster.

well-chosen starting point is enough for completion. Finally, we demonstrate that our method, LiDPM, leads to better results in scene completion on SemanticKITTI. The project page is https://astra-vision.github.io/LiDPM . [4/4 of https://arxiv.org/abs/2504.17791v1]

Tetiana Martyniuk, Gilles Puy, Alexandre Boulch, Renaud Marlet, Raoul de Charette
LiDPM: Rethinking Point Diffusion for Lidar Scene Completion
https://arxiv.org/abs/2504.17791

Tetiana Martyniuk, Gilles Puy, Alexandre Boulch, Renaud Marlet, Raoul de Charette: LiDPM: Rethinking Point Diffusion for Lidar Scene Completion https://arxiv.org/abs/2504.17791 https://arxiv.org/pdf/2504.17791 https://arxiv.org/html/2504.17791

LiDPM is a collaboration between Inria and @valeoai.bsky.social, couldn't have done it without the support of my amazing supervisors @alexandreboulch.bsky.social, @gillespuy.bsky.social, Renaud Marlet and Raoul de Charette!

We show that "vanilla" diffusion can be applied to large-scale (here - lidar) non-normalized point clouds.

LiDPM
● does not require any additional regularization,
● inherently extends to unconditional scene generation,
● outperforms "local" diffusion for scene completion on SemanticKITTI.