Paper Image

Learning unified vision representations for autonomous driving

Published on:

8 September 2023

Primary Category:

Computer Vision and Pattern Recognition

Paper Authors:

Thomas E. Huang,

Yifan Liu,

Luc Van Gool,

Fisher Yu

Bullets

Key Details

A new challenge with 10 diverse vision tasks is proposed to study multi-task learning for autonomous driving

Tasks include classification, segmentation, detection, pose, optical flow, and multi-object tracking

A network is designed to share representations across all tasks using progressive training and feature interaction

The unified network improves over single-task models in accuracy and computation

AI generated summary

Learning unified vision representations for autonomous driving

This paper introduces a new challenge with ten visual tasks to explore learning a shared representation for major image and video recognition tasks needed for autonomous driving. The tasks span object classification, segmentation, localization, and tracking. A network is proposed to tackle all tasks jointly with a single set of weights, using techniques like progressive training and feature sharing modules. Experiments show the unified network can surpass specialized single-task models in accuracy and efficiency.

Answers from this paper

Comments

No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up