Can I read A Multi-Task Model for Sea-Sky Scene Perception with Information Intersection on EtoBox?

A Multi-Task Model for Sea-Sky Scene Perception with Information Intersection by Qiang Wu; Qiyong Yang; Xin Zheng is a scholarly article available to read on EtoBox.

What is A Multi-Task Model for Sea-Sky Scene Perception with Information Intersection about?

Sea-sky scene is one of the most commonly used scenes of automatic perception technology. An accurate and comprehensive perceptive system, which mainly includes sea surface object detection and sea-sky line positioning tasks, can provide effective assistance to the marine rescue and defense. However, existing related single-task methods only focused on one of them separately. Accomplish the entire perception task by multiple algorithms will cause instability and redundancy of the system. In this paper, we proposed a Multi-Task Model for Sea-Sky Scene Perception which contains several parts to complete both of sub-tasks through one end-to-end inference. Our sea-sky line positioning part is built on the Anchor Classification based lane detection method. The sea surface object detection part is similar to the general detector called Generalized Focal Loss V2 (GFLV2) with a customize task-level information intersection module. Furthermore, by introducing information interaction at feature-level, a shared extractor through end-to-end training, our model achieves a promising performance in both latency and precision. Its superiority has been proved by the comparison result on a self-coll

Author

Qiang Wu; Qiyong Yang; Xin Zheng

Publisher

ACM

Published

2022

Language

EN