A Brief Overview of entire Project
- We found the most mundane problem in the world and decided to find a techy solution for it. Imagine you wake up every day and you have one less decision to make
What it does
- It takes your inventory of clothes in account with object category and attribute detection and gives you a proper everyday outfit recommendation based on what's going on in the market. On a positive side, a bad fashion sense will no more be a problem.
How we built it
Challenges we ran into
Training of a huge Inception Network on a dataset was near impossible on any hardware - Virtual or local. A lot of optimization and concepts of dimension reduction were used, not easy to implement. Pipeline connections to the website, cloud, and local machine were sensitive to changes and program fell apart real fast at several instances. A lot of hit and try methods in coming up with new Algorithms was indeed challenging and yet super fun.
Accomplishments that we're proud of-
Successfully completing a research level project in a short period of time including but not limited to coming up with a new algorithm of discriminator network that generated embeddings in a way that the general negative fashion style match was mapped far while trying to bring the positive matches near to each other.
A Step Beyond
The clothes that do not make the cut in users’ wardrobe are suggested to be donated a local homeless welfare organization.
What we learned
Implementations of Tensorflow and machine learning Libraries in different environments. Wix Development.
What's next for Fashion Sense
We plan on improving the accuracy of the network with proper training environments. We plan to introduce new features of Fashion recommendation based on weather, time, and other aspects that influence outfits of individuals. We plan to deploy the Homeless feature in an efficient way which can have a positive impact on the society. Also, we understand there always has been a difference in fashion styles among different genders, age groups, and other factors. We plan to implement it more user-friendly and user-specific that would only take in regard features that an individual is involved with.