ECCV 2022 Tutorial on

Self-Supervision on Wheels:
Advances in Self-Supervised Learning from Autonomous Driving Data

Monday, October 24 2022, 09:00 - 13:00 Tel Aviv time
(13:00 - 17:00 CST | 08:00 - 12:00 CET | 02:00 - 06:00 EDT)


The tremendous progress of deep-learning-based approaches to image understanding problems has inspired new advanced perception functionalities for autonomous systems. However, real-world vision applications often require models that can learn from large bulks of unlabeled and uncurated data with few labeled samples, usually costly to select and annotate. In contrast, typical supervised methods require extensive collections of carefully selected labeled data, a condition that is seldom fulfilled in practical applications. Self-supervised learning (SSL) arises as a promising line of research to mitigate this gap by training models using various supervision signals extracted from the data itself, without any human-generated labels.

SSL has seen a lot of exciting progress in the last two years, with many new SSL methods managing to match or even surpass the performance of fully supervised techniques. While most popular SSL methods revolve around web image datasets, e.g., ImageNet, new diverse forms of self-supervision are investigated for autonomous driving (AD). AD represents a unique sandbox for SSL methods as it brings among the largest public data collections in the community and provides some of the most challenging computer vision tasks: object detection, depth estimation, image-based odometry and localization, etc. Here, the canonical SSL pipeline (i.e., self-supervised pre-training a model and fine-tuning it on a downstream task) is revisited and extended to learn tasks for which ground-truth annotations are difficult to compute (e.g., dense depth) leading to utterly new SSL approaches for computer vision and robotics. In this tutorial we will provide an in-depth coverage of the various paradigms for self-supervised learning (old and new) through the lens of essential perception tasks for AD. Specifically, the tutorial will cover the following subjects: (1) Self-supervised representation learning from autonomous driving data, (2) Self-supervised learning for depth estimation, (3) Self-supervised learning for 3D detection and tracking, (4) Self-supervised learning for odometry and localization.


09:00 - 09:10 Tel Aviv time (08:00 - 08:10 CET) . Introduction by Patrick

09:10 - 10:00 Tel Aviv time (08:10 - 09:00 CET) . Self-supervised representation learning from AD data by Spyros and Andrei [slides]

10:05 - 10:55 Tel Aviv time (09:05 - 09:55 CET) . Self-supervised learning for depth estimation by Vito and Adrien [slides]

11:00 - 11:50 Tel Aviv time (10:00 - 10:50 CET) . Self-supervised learning for 3d tracking and detection by Adam and Katerina [slides]

11:55 - 12:45 Tel Aviv time (10:55 - 10:45 CET) . Self-supervised learning for odometry and localization by Daniele and Paul [slides]

12:50 - 13:00 Tel Aviv time (11:50 - 12:00 CET) . Conclusion by All

Please contact Spyros Gidaris if you have questions.

This webpage template is by courtesy of Georgia Gkioxari.

Last updated: 30 July 2022