How good are detection proposals, really?

TL;DR

This paper critically evaluates the effectiveness of various object detection proposal methods, analyzing their recall, repeatability, and impact on detector performance to inform better selection for different applications.

Contribution

It provides a comprehensive analysis of ten object proposal methods, highlighting their strengths and weaknesses to guide future improvements and applications.

Findings

Existing methods have notable weaknesses in recall and repeatability.

Detection proposals significantly influence detector performance.

Insights help select appropriate methods for specific settings.

Abstract

Current top performing Pascal VOC object detectors employ detection proposals to guide the search for objects thereby avoiding exhaustive sliding window search across images. Despite the popularity of detection proposals, it is unclear which trade-offs are made when using them during object detection. We provide an in depth analysis of ten object proposal methods along with four baselines regarding ground truth annotation recall (on Pascal VOC 2007 and ImageNet 2013), repeatability, and impact on DPM detector performance. Our findings show common weaknesses of existing methods, and provide insights to choose the most adequate method for different settings.