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

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.

Determining the proper number of proposals for individual images

“…In order to accurately localise the region proposals generated by the RPN, some methods [14][15][16][17] are proposed that perform multi-stage refinements that take the output of the previous stage as the input of the next stage till the accurate localization is obtained. Zhong [14] proposed the Iterative RPN that contains two sequential RPNs; the outputs smaller than 64 2 from the first RPN and larger than 64 2 from the second RPN were adopted according to the experiment.…”

“…Vu [16] proposed Cascade RPN that designed the adaptive convolutional layer which was employed at the second stage; it convolved the corners, centre points, and midpoints of the four sides of the proposals given by the former stage to make the extracted feature aligned with the corresponding proposal. Zou [17] proposed SC‐RPN that designed a lightweight IoU‐Mask branch to prevent high‐quality region proposals from being filtered at the first stage, and referred Reppoints [18] to produce 9 offsets for grouping a proposal with stronger fitting ability at the second stage.…”