Inspiration
We were inspired by NASA's use of machine learning used to detect exoplanets within out galaxy. Their model worked by measuring the brightness of stars far away from earth. When these stars were dimmed on a consistent basis, they knew they had discovered a planet. While our model may not be as complex as NASA's, it uses the same guiding principles.
What it does
Planeta is an app designed to help individuals learn about and experience astrology. Planeta uses machine learning to detect planets from pictures and videos in real time, and gives information about these planets to the user. The current model was trained using planets from our solar system, but future models could include planets from far beyond.
How we built it
We built Planeta using the imageAI library. Creating a algorithm to train the machine learning model wasn't the difficult part. The difficult part of the process was creating the planet data set, as there was no widely available one online. This took an all nighter from the both of us, as compiling a data set isn't a task meant for a single day. In the end, we were able to download and assign values to over 1100 images, and had some time left over to train the model.
Challenges we ran into
We ran into plenty of challenges while creating Planeta. The first problem was our webcam input. We couldn't get the openCV library to detect our web cameras. Ultimately, we solved the issue by setting our web cameras as the default device in device manager.
Later on in the process, we had difficulty when trying to convert the model from only being able to identify objects in pictures, to being able to identify objects in videos. We fixed this my reading through the openAI documentation, and learning about all the possible uses of it.
Finally, we ran into A LOT of issues while trying to compile our data set. Many of the images we downloaded from google ended up breaking the program. It took us hours to find out the exact files that were causing our program to fail. For anybody tryin this in the future... do NOT download .gifs or .webps, they don't work.
Accomplishments that we're proud of
We're extremely proud of the fact that we were able to compile a brand new data set of images that was not previously available on the internet. We did this with zero knowledge of how to compile a data set beforehand, and learned a lot as a result of our efforts.
What we learned
We learned a bunch about how machine learning works, and how training a set of data works. Specifically, we learned all the steps needed to prepare an image to be understood by the computer, such as outlining the object in a .xmp files, and creating validation folders for the model to test itself against.
What's next for Planeta
We have a bunch of cool ideas for Planeta. Firstly, we need to train the model for longer. One of the major downsides to short term machine learning is the accuracy isn't always great. Additionally we want to add support for star constellations and dwarf planets. We're also interested in the possibility of offering more info about the planets for advanced users.
Log in or sign up for Devpost to join the conversation.