Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

LidarCLIP or : How i Learned to Talk to Point Clouds

Hess, Georg ; Tonderski, Adam LU orcid ; Petersson, Christoffer ; Astrom, Kalle LU orcid and Svensson, Lennart (2024) 2024 IEEE Winter Conference on Applications of Computer Vision, WACV 2024 p.7423-7432
Abstract

Research connecting text and images has recently seen several breakthroughs, with models like CLIP, DALL•E 2, and Stable Diffusion. However, the connection between text and other visual modalities, such as lidar data, has received less attention, prohibited by the lack of text-lidar datasets. In this work, we propose LidarCLIP, a mapping from automotive point clouds to a pre-existing CLIP embedding space. Using image-lidar pairs, we supervise a point cloud encoder with the image CLIP embeddings, effectively relating text and lidar data with the image domain as an intermediary. We show the effectiveness of Lidar-CLIP by demonstrating that lidar-based retrieval is generally on par with image-based retrieval, but with complementary... (More)

Research connecting text and images has recently seen several breakthroughs, with models like CLIP, DALL•E 2, and Stable Diffusion. However, the connection between text and other visual modalities, such as lidar data, has received less attention, prohibited by the lack of text-lidar datasets. In this work, we propose LidarCLIP, a mapping from automotive point clouds to a pre-existing CLIP embedding space. Using image-lidar pairs, we supervise a point cloud encoder with the image CLIP embeddings, effectively relating text and lidar data with the image domain as an intermediary. We show the effectiveness of Lidar-CLIP by demonstrating that lidar-based retrieval is generally on par with image-based retrieval, but with complementary strengths and weaknesses. By combining image and lidar features, we improve upon both single-modality methods and enable a targeted search for challenging detection scenarios under adverse sensor conditions. We also explore zero-shot classification and show that LidarCLIP outperforms existing attempts to use CLIP for point clouds by a large margin. Finally, we leverage our compatibility with CLIP to explore a range of applications, such as point cloud captioning and lidar-to-image generation, without any additional training. Code and pre-trained models at github.com/atonderski/lidarclip.

(Less)
Please use this url to cite or link to this publication:
author
; ; ; and
organization
publishing date
type
Chapter in Book/Report/Conference proceeding
publication status
published
subject
keywords
Algorithms, Applications, Autonomous Driving, Vision + language and/or other modalities
host publication
Proceedings - 2024 IEEE Winter Conference on Applications of Computer Vision, WACV 2024
pages
10 pages
publisher
IEEE - Institute of Electrical and Electronics Engineers Inc.
conference name
2024 IEEE Winter Conference on Applications of Computer Vision, WACV 2024
conference location
Waikoloa, United States
conference dates
2024-01-04 - 2024-01-08
external identifiers
  • scopus:85188146235
ISBN
9798350318920
DOI
10.1109/WACV57701.2024.00727
language
English
LU publication?
yes
additional info
Publisher Copyright: © 2024 IEEE.
id
7c9f8308-87df-43af-998b-48dc4a30b4d2
date added to LUP
2024-06-07 14:24:28
date last changed
2024-08-12 16:44:05
@inproceedings{7c9f8308-87df-43af-998b-48dc4a30b4d2,
  abstract     = {{<p>Research connecting text and images has recently seen several breakthroughs, with models like CLIP, DALL•E 2, and Stable Diffusion. However, the connection between text and other visual modalities, such as lidar data, has received less attention, prohibited by the lack of text-lidar datasets. In this work, we propose LidarCLIP, a mapping from automotive point clouds to a pre-existing CLIP embedding space. Using image-lidar pairs, we supervise a point cloud encoder with the image CLIP embeddings, effectively relating text and lidar data with the image domain as an intermediary. We show the effectiveness of Lidar-CLIP by demonstrating that lidar-based retrieval is generally on par with image-based retrieval, but with complementary strengths and weaknesses. By combining image and lidar features, we improve upon both single-modality methods and enable a targeted search for challenging detection scenarios under adverse sensor conditions. We also explore zero-shot classification and show that LidarCLIP outperforms existing attempts to use CLIP for point clouds by a large margin. Finally, we leverage our compatibility with CLIP to explore a range of applications, such as point cloud captioning and lidar-to-image generation, without any additional training. Code and pre-trained models at github.com/atonderski/lidarclip.</p>}},
  author       = {{Hess, Georg and Tonderski, Adam and Petersson, Christoffer and Astrom, Kalle and Svensson, Lennart}},
  booktitle    = {{Proceedings - 2024 IEEE Winter Conference on Applications of Computer Vision, WACV 2024}},
  isbn         = {{9798350318920}},
  keywords     = {{Algorithms; Applications; Autonomous Driving; Vision + language and/or other modalities}},
  language     = {{eng}},
  pages        = {{7423--7432}},
  publisher    = {{IEEE - Institute of Electrical and Electronics Engineers Inc.}},
  title        = {{LidarCLIP or : How i Learned to Talk to Point Clouds}},
  url          = {{http://dx.doi.org/10.1109/WACV57701.2024.00727}},
  doi          = {{10.1109/WACV57701.2024.00727}},
  year         = {{2024}},
}