Publication Date

Spring 2019

Degree Type

Master's Project

Degree Name

Master of Science (MS)

Department

Computer Science

First Advisor

Robert Chun

Second Advisor

Thomas Austin

Third Advisor

Kevin Smith

Keywords

Convolutional neural networks, deep learning, Faster R-CNN, neural networks, object detection, SSD, YOLO

Abstract

Object detection is a very challenging problem in computer vision and has been a prominent subject of research for nearly three decades. There has been a promising in- crease in the accuracy and performance of object detectors ever since deep convolutional networks (CNN) were introduced. CNNs can be trained on large datasets made of high resolution images without flattening them, thereby using the spatial information. Their superior learning ability also makes them ideal for image classification and object de- tection tasks. Unfortunately, this power comes at the big cost of compute and memory. For instance, the Faster R-CNN detector required 180 billion FLOPs for training, and has over 100 million parameters.

In this project, we explore the popular state-of-the-art object detectors and present their contributions and shortcomings. Then we explore the recent lightweight detectors which try to address the issue of high resource requirements by building leaner models. Building upon the contributions of the state-of-the-art object detectors, and recent de- velopments in CNN training, we propose our own lightweight detector. We proposed a novel CNN block, to improve the inter-channel dependency in feature maps, called the inter-channel dependency block (ICDB). Through experiments on benchmark datasets we demonstrated our model attains better accuracy compared to the previous methods. Three benchmarking datasets PASCAL VOC 2007, KITTI and COCO have been used to demonstrate that our model scales well to different scenarios.

Share

COinS