Image recognition is one of the main branches of computer vision in Artificial Intelligence (AI), and one of the many things that I had the opportunity to learn during my internship here at Wovenware this past summer.
My experience is something that I’m confident will serve me greatly as I begin my career, and also something that I will cherish forever.
As a student at the University of Puerto Rico in Mayagüez without any prior internship experience or expertise in AI, frankly, I was terrified of what would be expected of me and everything was new and overwhelming at first.
Nevertheless, hope was not lost, and with each passing day I worked hard to learn as much as possible so I would be able to complete my assigned tasks, which included assisting with the analysis and processing of data that would go into the creation of machine learning models for image recognition.
Building a Machine Learning Model to Recognize Images
Image recognition is the ability for software to identify any type of object or living thing inside a picture. Our goal was to create a model that could identify and predict the exact location of a variety of objects like buildings, aircrafts, and vehicles from a given picture. The model consists of a method named SSD, which is the process of predicting the location of the object inside a bounding box, a four point coordinate that encloses the predicted object of the desired category.
To handle our data, we used the pandas software library, written for Python, which helped us create a data frame to control, use and extract the desired data in a very few lines of code. The other part of this project required us to train and validate the models we created using a class with a factory design pattern.
The essential part of our project involved a single class that followed this pattern, to generated similar objects without the use of constructors. In this process, the factory receives all of the data that the client needs and it pulls all of the required classes to build that object with the client’s specifications. One of my main tasks was to write a code for reading the configurations of the desired model and generate it in a YAML file. Our model was generated by this factory class, and while it was being created, it went through the process of training in a deep neural network using the Single Shot Detector (SSD).
Thanks to this internship, I was able to expand my knowledge of fascinating news areas of AI, and work hands on with models just as a professional data scientist. I’m grateful for the opportunity not only professionally and academically, but also for the relationships I have created with my co-workers, and the opportunity to work at a dynamic and caring company. My supervisor, Leslie De Jesus and mentor Brian Landron were determined to teach me everything about the project, and helped me learn as much as I could so that I have the tools needed to continue on my own.
As I make my way to Google for a semester co-op, I will never forget Wovenware, nor the friendships and memories I’ve made here. They say that “you’ll never forget your first,” like your first car, your first love, or your first job. Well, I will certainly never forget my first internship.