Enhance the recognition ability to occlusions and small objects with Robust Faster R-CNN

Abstract

Recognizing objects with vastly different size scales and objects with occlusions is a fundamental challenge in computer vision. This paper addresses this issue by proposing a novel approach denoted as Robust Faster R-CNN for detecting objects in multi-label images. Robust Faster R-CNN employs a cascaded network structure based on the Faster R-CNN architecture to extract features from objects with different size scales. However, the proposed design provides greater robustness than Faster R-CNN by replacing the RoIPooling operation with RoIAligns to eliminate the harsh quantization conducted by RoIPooling, and we design a multi-scale RoIAligns operation by adding multiple pool sizes for adapting the detection ability of the network to objects with different sizes. Furthermore, we combine an adversarial network with the proposed network to generate training samples with occlusions significantly affecting the classification ability of the model, which improves its robustness to occlusions. Experimental results for the PASCAL VOC 2012 and 2007 datasets demonstrate the superiority of the proposed object detection approach relative to several state-of-the-art approaches.