ReSTiNet: On Improving the Performance of Tiny-YOLO-Based CNN Architecture for Applications in Human Detection

Human detection is a special application of object recognition and considered one the greatest challenges in computer vision. It starting point number applications, including public safety security surveillance around world. technologies have advanced significantly recent years due to rapid development deep learning techniques. Despite advances, we still need adopt best network-design practices that enable compact sizes, designs, fast training times while maintaining high accuracies. In this article, propose ReSTiNet, novel compressed convolutional neural network addresses issues size, speed, accuracy. Following SqueezeNet, ReSTiNet adopts fire modules by examining their placement within model reduce parameters thus size. The residual connections are interpolated finely constructed improve feature propagation ensure largest possible information flow model, with goal further improving proposed terms speed algorithm downsizes previously popular Tiny-YOLO improves following features: (1) faster speed; (2) size; (3) solving overfitting problems; (4) superior performance than other lightweight models such as MobileNet SqueezeNet mAP. was trained tested using MS COCO Pascal VOC datasets. resulting 10.7 MB size (almost five smaller Tiny-YOLO), but it achieves an mAP 63.74% on PASCAL 27.3% datasets Tesla k80 GPU.