Inspiration
While in the search for an FLL project related to arts, one of our team members told us about their problem with finding that one specific photo in the pile of hundreds. We thought that this would be very beneficial not just for the photographers, but also for anyone with a camera.
After talking to more professionals we found out that this was a common issue among them. A program that could do the most time consuming part for them would leave the photographers with more time to be productive.
What it does
Photoshelf lets you search keywords and give the photos that can be defined by those. Let’s say that you are a car spotter. To find that bright red Ferrari you took a picture of 3 days ago you just have to search red Ferrari. The program will reduce the amount of pictures you have to deal with by only leaving you the red Ferrari’s.
How we built it
We used python for our project as it was the best choice. Before switching to our current model, we tried to train our own but unfortunately we didn’t have the time and resources required to create such a huge model so we decided to use OWL-ViT instead. As their huggingface page suggests: “OWL-ViT is a zero-shot text-conditioned object detection model.” It was the perfect fit for our project.
For ui we first tried to use tkinter but didn’t look as good and would be a waste of time if we tried to improve the looks. Streamlit was our second and final choice. Even though it is used to create websites, using streamlit for ui was also possible.
Challenges we ran into
Our biggest challenge during the process was deciding on what method to use. By using OWL-ViT we leaped miles ahead in terms of our progress. Other than that since it was our first time using this model we encountered issues related to installment of the library(for example some folders’ paths were too long) and its usage. With the help of an experienced individual in the area of artificial intelligence named Alihan Karadağ we got over these issues easily and learned how to prevent them later on.
Accomplishments that we're proud of
Our team’s biggest accomplishment was learning about algorithms, artificial intelligence and image recognition. Some of us had no programming background whatsoever but they were able to and teach about our project to others. They adapted very well and were quick learners.
What we learned
As mentioned above we learned about artificial intellgence and image recognition. Alihan Karadağ gave us literal lessons about how training a model works. We also learned how to code better in the process. Like writing more optimized programs and improving the understandibility of our codes.
What's next for PhotoShelf
PhotoShelf is a project that has many ways to improve in. It could be adapted to other devices like phones or further optimized to work faster and use less resources. The working prototype could be made more user-friendly and published, sharing our great solution to our important problem to the entire world.
Log in or sign up for Devpost to join the conversation.