FRFB: Integrate Receptive Field Block Into Feature Fusion Net for Single Shot Multibox Detector

SSD (Single Shot Multibox Detector) is one of the best object detection algorithms with both high accuracy and fast speed. FSSD (Feature Fusion Single Shot Multibox Detector) proposed feature fusion module which can improve the performance significantly. RFB Net(Receptive Field Block Net for Accurate and Fast Object Detection) proposed RFB module to simulate Receptive Fields (RFs) in human visual systems and gain higher accuracy. In this paper, we proposed FRFB Net (Integrate Receptive Field Block Feature into Fusion Net for Single Shot Multibox Detector), an enhanced FSSD with a RFB module,which not only fully utilize the pyramidal features, but also change the RFs of the fused feature map. To make the model more robust,we use Gaussian Blur to process training images,in addition to use the data augmentation in SSD.On the Pascal VOC 2007 test, our network can achieve 79.6 mAP with the input size $300\times 300$ using a single Nvidia 1080 GPU with any bells and whistles. In addition, our result on COCO is also better than FSSD, achieves 2.7mAP improvement compared to FSSD. Our FRFBNet outperforms a lot of state-of-the-art object detection algorithms in accuracy and speed.