Anchorless object detection with limited hardware resources
Supervisor:
Dr. Harmati István
External supervisor:
Deák-Meszlényi Regina, Continental
Subject:
Project Laboratory - Control systems study specialization, BSc Elec.
Students count:
1
Continue:
Szakdolgozat / Diplomaterv
Description:
Several new papers propose solutions for anchorboxless object detection to overcome the complexity of anchorbox parameter setup in state-of-the-art single-shot object detectors. One of the most promising architectures is CornerNet (https://arxiv.org/pdf/1808.
The task of the candidate is to test the original CornerNet architecture and check the feasibility of this anchorless approach with several mobile-sized backbone architectures e.g. MobileNetv3.
