as a group of friends, we often look at each other's Instagrams and roasting each other's sense of fashion for fun. however, manually reviewing each post is inefficient because there are so many Instagram posts. we get the most 'fun' in the most efficient manner, we created a neural net to automate this process.
What it does
you upload a fit pic, and our neural net roasts you accordingly.
How we built it
we use YOLO to find and crop out person-bounding boxes, then feed that cropped image into a deep neural network we trained based on five categories using TensorFlow. the neural network will respond with a prediction of how similar your fashion sense is to the five categories. it then uses this information to insult you.
we used React to build the front end and Flask to host our backend.
Challenges we ran into
in order to get more accurate results, we tried to do transfer training using resent50 as a backbone to improve the complexity of our model. we struggled to get the dimensions of input and output layers to match with our image input, outputs.
Accomplishments that we're proud of
we got it working without our uber eats credits to help us power through the night sadge
What we learned
half of our team had no prior experience with neural networks!
What's next for judge my fashion
next we will add more classes for the model to classify and come up with more insults!!!!