ProbaNet: Proposal-balanced Network for Object Detection
This addresses performance issues for users of two-stage object detectors, but it is incremental as it builds on existing methods with minor modifications.
The paper tackles the easy-hard sample imbalance problem in object detection by proposing ProbaNet, which discards easy samples and emphasizes foreground proposals, resulting in a 1.2% higher mAP on PASCAL VOC 2007 compared to the baseline.
Candidate object proposals generated by object detectors based on convolutional neural network (CNN) encounter easy-hard samples imbalance problem, which can affect overall performance. In this study, we propose a Proposal-balanced Network (ProbaNet) for alleviating the imbalance problem. Firstly, ProbaNet increases the probability of choosing hard samples for training by discarding easy samples through threshold truncation. Secondly, ProbaNet emphasizes foreground proposals by increasing their weights. To evaluate the effectiveness of ProbaNet, we train models based on different benchmarks. Mean Average Precision (mAP) of the model using ProbaNet achieves 1.2$\%$ higher than the baseline on PASCAL VOC 2007. Furthermore, it is compatible with existing two-stage detectors and offers a very small amount of additional computational cost.