Training Region-based Object Detectors with Online Hard Example Mining

TL;DR

This paper introduces an online hard example mining (OHEM) algorithm that improves training efficiency and detection accuracy for region-based ConvNet object detectors by automatically focusing on challenging examples.

Contribution

It proposes a simple, hyperparameter-free OHEM method that enhances detection performance and reduces reliance on heuristics, achieving state-of-the-art results on major benchmarks.

Findings

OHEM significantly boosts detection accuracy on PASCAL VOC datasets.

The method scales well with larger, more complex datasets like MS COCO.

OHEM achieves state-of-the-art mAP scores on PASCAL VOC 2007 and 2012.

Abstract

The field of object detection has made significant advances riding on the wave of region-based ConvNets, but their training procedure still includes many heuristics and hyperparameters that are costly to tune. We present a simple yet surprisingly effective online hard example mining (OHEM) algorithm for training region-based ConvNet detectors. Our motivation is the same as it has always been -- detection datasets contain an overwhelming number of easy examples and a small number of hard examples. Automatic selection of these hard examples can make training more effective and efficient. OHEM is a simple and intuitive algorithm that eliminates several heuristics and hyperparameters in common use. But more importantly, it yields consistent and significant boosts in detection performance on benchmarks like PASCAL VOC 2007 and 2012. Its effectiveness increases as datasets become larger and more difficult, as demonstrated by the results on the MS COCO dataset. Moreover, combined with complementary advances in the field, OHEM leads to state-of-the-art results of 78.9% and 76.3% mAP on PASCAL VOC 2007 and 2012 respectively.