Context-Aware Single-Shot Detector

TL;DR

This paper introduces CSSD, an enhancement of SSD that incorporates multi-scale context modeling through dilated and deconvolution layers, significantly improving small object detection accuracy while maintaining real-time performance.

Contribution

The paper proposes CSSD, a novel context-aware extension of SSD, with two variants using dilated and deconvolution layers, improving detection accuracy especially for small objects.

Findings

CSSD achieves +3.2% AP for small objects on MS-COCO.

Multi-scale context modeling improves overall detection accuracy.

CSSD maintains comparable runtime to original SSD.

Abstract

SSD is one of the state-of-the-art object detection algorithms, and it combines high detection accuracy with real-time speed. However, it is widely recognized that SSD is less accurate in detecting small objects compared to large objects, because it ignores the context from outside the proposal boxes. In this paper, we present CSSD--a shorthand for context-aware single-shot multibox object detector. CSSD is built on top of SSD, with additional layers modeling multi-scale contexts. We describe two variants of CSSD, which differ in their context layers, using dilated convolution layers (DiCSSD) and deconvolution layers (DeCSSD) respectively. The experimental results show that the multi-scale context modeling significantly improves the detection accuracy. In addition, we study the relationship between effective receptive fields (ERFs) and the theoretical receptive fields (TRFs), particularly on a VGGNet. The empirical results further strengthen our conclusion that SSD coupled with context layers achieves better detection results especially for small objects ($+3.2\% {\rm AP}_{@0.5}$ on MS-COCO compared to the newest SSD), while maintaining comparable runtime performance.