IoU-aware single-stage object detector for accurate localization

Abstract

Single-stage object detectors have been widely applied in many computer vision applications due to their simpleness and high efficiency. However, the low correlation between the classification score and localization accuracy in detection results severely hurts the average precision of the detection model. To solve this problem, an IoU-aware single-stage object detector is proposed in this paper. Specifically, IoU-aware single-stage object detector predicts the IoU for each detected box. Then the predicted IoU is multiplied by the classification score to compute the final detection confidence, which is more correlated with the localization accuracy. The detection confidence is then used as the input of the subsequent NMS and COCO AP computation, which substantially improves the localization accuracy of model. Sufficient experiments on COCO and PASCOL VOC datasets demonstrate the effectiveness of IoU-aware single-stage object detector on improving model's localization accuracy. Without whistles and bells, the proposed method can substantially improve AP by 1.7%–1.9% and AP75 by 2.2%–2.5% on COCO test-dev. And it can also substantially improve AP by 2.9%–4.4% and AP80, AP90 by 4.6%–10.2% on PASCAL VOC. The source code will be made publicly available.