In this project I tried to predict skin cancer type from one or more images, combined with some extra information about the patient: age, sex, and the location on the body.
The base model is mobilenetv3_small_100. This choice decreases accuracy and precision a bit compared to heavier backbones, but the main goal of this project was to keep processing time very small, even on weak devices. So I tried to find a trade-off between resources, time, and accuracy, rather than just chasing the highest possible score.
Maybe someday I'll do the deployment part myself, by building a phone app that installs and runs the model locally on your device.
I appreciate any contribution to the progress of this project.