I came into this program with no idea what I was going to be doing and definitely no experience with machine learning. As the deadline for the final poster presentation approaches and I formulate my project, I realize that I have learned so much since June. The techniques that I have been learning for the past six weeks have been introduced to me so that I will hopefully be able to implement these ideas.
At the beginning of the week, I only knew that my final project aims to mount a Jetson and camera on a drone to classify objects in real-time. Therefore, I naturally tried to get the Jetson up and running. Unfortunately, Chris and I ran into a lot of problems with powering up the Jetson. The power light turned on easily, but nothing ever displayed on the monitor, causing us to think that there is not enough current running through the Jetson when additional hardware (monitor, mouse, keyboard, etc.) are connected. We ordered a new power supply cable and are waiting for the package.
In order to learn more about the software of the project, Chris and I then started working on categorizing images of 3-dimensional objects. We looked up online tutorials for Convolutional Neural Networks (CNNs) and created a sign language classifier! There were 16 more classes than before (because there are 26 letters and only 10 digits), and it was really exciting when the accuracy got to around 90%.
Next, it was time to really start formulating a project. After talking with Ji Young and Dr. Detweiler, Chris and I decided to do similar projects dealing with the detection of drones. However, the projects are slightly different because he wants to look into regularization techniques while I want to look into the effect of detection with an increased number of classes. In other words, I want too add airplanes, birds, and helicopters to the dataset of drones and see if the Jetson can distinguish these objects.
After having formulated the project, Chris and I have been working on training the computer to recognize drones. We were very lucky to come upon already-labeled data, but I have been running into problems with Google Colab once again. From directory paths to not being able to view images in a pop-up window, these problems seem to be a normality now, and I am getting better at debugging the problems.