(2024.06.25 ~ 2025.01.22)
This is an international STEM competition that allows students to present and display their creativity through long term projects in their own criterias. I collaborated with two friends majoring in computer science , and I acted as a chemistry assistant.
Our main project was about creating an AI that could detect glass apparatus all at once. We aimed to provide a solution for factories or science laboratories that are often unorganized, helping them process and manage equipment more efficiently and quickly.
Before we started doing things, I had a lot of things to set. I prepared various types of glass apparatuses that were needed for the project, making sure diverse examples were available for testing and validation
Just as shown in the picture, we took pictures of the glass apparatuses in various angles and locations, to train the AI to adapt best to any kind of environment. As the training quantity increases, the more accurate and meticulous the AI is trained.
After we took tons of different pictures,we trained them over several times again and again to make it ready. The blue line that is shown on the image is indicating the classification and validity. The more the number is closer to 1, the more it’s accurate.
-> This graph is one of the training processes that contains four major types of losses: box losses, classification losses, training losses, and validation losses. Also one of the major focuses I had for this project. In summary, if the model shows both training and validation losses decreasing over time, this means that the model is learning not just on the training data but also across different datasets to detect and classify objects effectively in more general settings. During this analysis, I learned that there were more factors that impacted the result than we thought. Thus, I had to study the impact of light reflection, transparency, and angle variations to understand how they affected Al recognition.
–> This is the paper we wrote that organizes our long project in a few pages.
( + updates : we got an bronze award for the domestic competition ! )
답글 남기기