ASSD: Attentive Single Shot Multibox Detector

TL;DR

ASSD introduces an attentive neural network for object detection that emphasizes relevant regions in feature maps, achieving competitive accuracy with simpler, more efficient design compared to existing methods.

Contribution

The paper presents ASSD, a novel attentive network that models feature relations spatially, improving detection accuracy while maintaining simplicity and computational efficiency.

Findings

ASSD performs favorably against state-of-the-art detectors.

It effectively highlights useful regions in feature maps.

The model is computationally efficient and simple in design.

Abstract

This paper proposes a new deep neural network for object detection. The proposed network, termed ASSD, builds feature relations in the spatial space of the feature map. With the global relation information, ASSD learns to highlight useful regions on the feature maps while suppressing the irrelevant information, thereby providing reliable guidance for object detection. Compared to methods that rely on complicated CNN layers to refine the feature maps, ASSD is simple in design and is computationally efficient. Experimental results show that ASSD competes favorably with the state-of-the-arts, including SSD, DSSD, FSSD and RetinaNet.