Bal-R$^2$CNN: High Quality Recurrent Object Detection With Balance Optimization

Xiaoyu Chen, Hongliang Li, Qingbo Wu, Fanman Meng, Heqian Qiu

Abstract: It is a common practice to refine object detection results using recurrent detection paradigm. We evaluate the recurrent detection on Faster R-CNN, but the improvement is far away from expected. We consider that the performance bottleneck is from <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">imbalance optimization</i> caused by the biased distribution of training data. Low-IoU-skewed RPN proposals could suppress the contribution of High-IoU examples at the training stage. Besides, data imbalance and statistical discrepancy on regression targets between low-IoU and high-IoU examples are not considered in the regression task; this design could impede localization quality. In this work, we propose Bal-R <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^2$</tex-math></inline-formula> CNN for high-quality recurrent object detection. There are two new components in Bal-R <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^2$</tex-math></inline-formula> CNN. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Self-iteration box sampling</i> collects object boxes from recurrent steps and increases the number of high-IoU training examples. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">IoU-sensitive bounding-box regression</i> sends proposal boxes with different IoUs to specified regression branches for more accurate bounding-box prediction. Both two new components could induce <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">balanced optimization</i> and be helpful. With the resulting Bal-R <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^2$</tex-math></inline-formula> CNN detector, evaluation on PASCAL VOC and MSCOCO reveal that our method has a significant improvement on the existing solution and could reach a better performance than several state-of-the-art methods.