SC-RPN: A Strong Correlation Learning Framework for Region Proposal

TLDR: Wang et al. as discussed by the authors proposed a two-stage strong correlation learning framework, abbreviated as SC-RPN, which aims to set up stronger relationship among different modules in the region proposal task.

Abstract: Current state-of-the-art two-stage detectors heavily rely on region proposals to guide the accurate detection for objects. In previous region proposal approaches, the interaction between different functional modules is correlated weakly, which limits or decreases the performance of region proposal approaches. In this paper, we propose a novel two-stage strong correlation learning framework, abbreviated as SC-RPN, which aims to set up stronger relationship among different modules in the region proposal task. Firstly, we propose a Light-weight IoU-Mask branch to predict intersection-over-union (IoU) mask and refine region classification scores as well, it is used to prevent high-quality region proposals from being filtered. Furthermore, a sampling strategy named Size-Aware Dynamic Sampling (SADS) is proposed to ensure sampling consistency between different stages. In addition, point-based representation is exploited to generate region proposals with stronger fitting ability. Without bells and whistles, SC-RPN achieves AR1000 14.5% higher than that of Region Proposal Network (RPN), surpassing all the existing region proposal approaches. We also integrate SC-RPN into Fast R-CNN and Faster R-CNN to test its effectiveness on object detection task, the experimental results achieve a gain of 3.2% and 3.8% in terms of mAP compared to the original ones.

Figures

Fig. 3. The generation of supervision information and the training phase of Light-weight IoU-Mask branch. As a result, red points in ground-truth IoU map denote the ground-truth of training samples, while white points denote the irrelevant samples.

Fig. 2. The training phase of SC-RPN with Size-Aware Dynamic Sampling (SADS). In the first stage, positive samples are output and all of them are regarded as regression samples. Then in the second stage, positive and negative samples are output, among which all the positive samples are regarded as regression samples, while all the positive samples and the negative samples after random sampling are regarded as classification samples.

TABLE II PERFORMANCE OF OBJECT DETECTORS ON MS COCO 2017 TEST-DEV SPLIT.

TABLE I PERFORMANCE OF REGION PROPOSAL APPROACHES ON MS COCO 2017 VAL SPLIT.

TABLE IV ABLATION ANALYSIS OF DIFFERENT SAMPLING STRATEGY COMBINATIONS IN SADS. POSR AND SAT DENOTE POSITIVE REGION ASSIGNER AND SIZE-AWARE THRESHOLD ASSIGNER, RESPECTIVELY.

TABLE III COMPONENT-WISE PERFORMANCE OF SC-RPN.