Small Object Detection using Context and Attention

TL;DR

This paper introduces a novel small object detection method that leverages multi-scale context features and attention mechanisms to enhance accuracy, outperforming conventional models like SSD on benchmarks such as PASCAL VOC2007.

Contribution

It proposes a new detection approach combining multi-scale context features and attention mechanisms specifically for small object detection.

Findings

Achieved 78.1% mAP on PASCAL VOC2007 with 300x300 input.

Outperformed conventional SSD in small object detection accuracy.

Demonstrated effectiveness of context and attention in improving detection results.

Abstract

There are many limitations applying object detection algorithm on various environments. Especially detecting small objects is still challenging because they have low resolution and limited information. We propose an object detection method using context for improving accuracy of detecting small objects. The proposed method uses additional features from different layers as context by concatenating multi-scale features. We also propose object detection with attention mechanism which can focus on the object in image, and it can include contextual information from target layer. Experimental results shows that proposed method also has higher accuracy than conventional SSD on detecting small objects. Also, for 300$\times$300 input, we achieved 78.1% Mean Average Precision (mAP) on the PASCAL VOC2007 test set.