Multiple spatial residual network for object detection

Abstract Many residual network-based methods have been proposed to perform object detection. However, most of them may lead overfitting or cannot well in small detection and alleviate the problem overfitting. We propose a multiple spatial network (MSRNet) for Particularly, our method is based on central point algorithm. Our MSRNet employs as backbone. The resulting features are processed by channel pooling module. then construct multi-scale feature transposed fusion structure consists three overlapping stacked convolution modules transpose function. Finally, we use Center process high-resolution image obtaining final prediction result. Experimental results PASCAL VOC dataset COCO confirm that has competitive accuracy compared with several other classical algorithms, while providing unified framework training reasoning. runs GeForce RTX 2080Ti.