Comparing YOLO and MixNet architectures for image-based human detection

Abstract

[en] Object detection is a technique that allows computers to identify objects in images or videos. The technique most commonly used for this operation is called Convolutional Neural Network (CNN), because of its good performance. Object detection had a big impact in the last two decades, because of its wide range of industries where it can be applied. Among which we can find autonomous driving where cars have to decide by their own when to accelerate, turn, brake… face detection which can be used for unlocking phones or surveillance among others, object extraction of images, personal identification through iris code, smile detection for cameras, medical image processing tools and many more. We see the importance of finding ways to improve the way we teach computers to understand images, so we can have autonomous machines that are more accurate and reliable. Our goal in this project is to study the performance of different architecture designs and techniques in the task of object detection. This thesis could help as a guide for future projects to observe how changes with data augmentation and different architecture designs can affect their model.