# Assisted Excitation of Activations: A Learning Technique to Improve Object Detectors

## TL;DR

This paper introduces a simple learning technique called assisted excitation that enhances YOLO object detectors' accuracy by guiding activation learning during training, achieving state-of-the-art speed-accuracy trade-offs.

## Contribution

It proposes a novel activation excitation method during training that improves object detection accuracy without affecting inference speed.

## Key findings

- YOLOv2 mAP increased by 3.8% on MSCOCO
- YOLOv3 mAP increased by 2.2% on MSCOCO
- Technique is simple, effective, and applicable to most single-stage detectors

## 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.