Assisted Excitation of Activations: A Learning Technique to Improve Object Detectors
Computer Vision and Pattern Recognition
2019-06-14 v1
Abstract
We present a simple and effective learning technique that significantly improves mAP of YOLO object detectors without compromising their speed. During network training, we carefully feed in localization information. We excite certain activations in order to help the network learn to better localize. In the later stages of training, we gradually reduce our assisted excitation to zero. We reached a new state-of-the-art in the speed-accuracy trade-off. Our technique improves the mAP of YOLOv2 by 3.8% and mAP of YOLOv3 by 2.2% on MSCOCO dataset.This technique is inspired from curriculum learning. It is simple and effective and it is applicable to most single-stage object detectors.
Keywords
Cite
@article{arxiv.1906.05388,
title = {Assisted Excitation of Activations: A Learning Technique to Improve Object Detectors},
author = {Mohammad Mahdi Derakhshani and Saeed Masoudnia and Amir Hossein Shaker and Omid Mersa and Mohammad Amin Sadeghi and Mohammad Rastegari and Babak N. Araabi},
journal= {arXiv preprint arXiv:1906.05388},
year = {2019}
}