Overview
Paisley Reimagined was my final project for my Artificial Intelligence course in spring 2023. The assignment asked us to pick a real-world AI application, research it properly, and then actually build it: implement a learning agent, test it, run experiments, and reflect on the ethics of the thing we made. Mine was a flower identifier, an object detection model trained to recognize types of flowers in photos.
The Pipeline
The project lives in a Colab notebook. The flower dataset sits in my Google Drive, sorted into one folder per flower type, and the notebook walks those folders with OpenCV, converting each image to grayscale and turning the folder names into training labels with scikit-learn. The model itself is EfficientDet-Lite0 through TensorFlow Lite Model Maker, trained with the whole model unfrozen, then evaluated and exported as a .tflite file so it could run on small devices.
The Test Subject
My favorite files in the project folder are the masking tests: a photo of Flint the beagle, mid tongue flop, next to versions of the same photo where everything around him is blacked out. Flowers are lovely, but nothing tells you an image pipeline is actually doing something like a dog with his tongue out.
What It Required
- A 500+ word proposal with at least three motivating references
- A working implementation of a learning agent, with testing to back it up
- Experiments showing how the application performed under different conditions
- A written reflection on the ethical considerations of the application
- A 2000+ word final report with reproducible setup instructions
We also demonstrated our projects live in the campus center lobby partway through, then used the feedback to finish before the deadline. Demoing an unfinished AI project to whoever walks by is a special kind of character building. The notebook I still have is honest about that stage of the work too: one cell has a comment that just says fix dimensional error, and I am keeping it that way.