A need to explore data science.
What it does
Allows a user to upload an image of an unknown celebrity to our website which gets process on the backend and presents predicted celebrity names with aesthetically formatted confidence intervals.
How we built it
Teamwork, splitting up into front end, back end, and ML workflows. We found a really nice Python API called face_recognition that uses a pre-trained deep neural network to process input images and produce encodings. After we created a library of celebrity face encodings by creating a python script that scrapes IMDb, we were able to take in new images and compare them to our existing encodings to output which celebrity the face corresponds with.
Challenges we ran into
We decided to go with a node backend using Jade, but we initially wanted to use react or a new framework that we didn't know. We found that this was too ambitious in the short time of a hackathon to do well.
Accomplishments that we're proud of
Our confidence interval algorithm is pretty interesting. We decided to use a threshold to determine which celebrities to include in the output, and then applied a reciprocal softmax algorithm to display the results to the user.