Inspiration
We're a group of 4 college students who each have struggled in creating a fun, stylistic wardrobe. From a combination of our fashion naivety and desire for easy decision-making, Komo was born to make our shopping experiences easier.
What it does
Upload pictures of your clothes and using AWS Rekognition services, our program will identify what you have and try to match it to an existing database of complete outfits from various company stores.
How we built it
Using AWS, we used a custom data set to train a AWS Rekognition model to be able to identify clothes and colors. Rekognition will label all the relevant items of clothing and we use that to match to existing outfits using the available clothes submitted by the user, which are also processed using the same technique.
Challenges we ran into
- We didn't have much experience with GUI development and had several roadblocks on how the frontend should work. Initially trying to make a simple web interface, we resorted to a simple Python form due to inexperience.
- The amount of time necessary for complete and accurate training of our AI was too volatile, with greater numbers varying in training time into full days, not to mention a significant amount of manual labeling.
Accomplishments that we're proud of
We are able to achieve the main goal of creating an outfit given a set of existing clothing!
What we learned
- We learned a lot about wed dev along the way despite not using it as our front end at the end as we struggled against the current for a significant portion of the remaining hours.
- Despite our interests in the topic, Komo represents our first introduction to Machine learning and the challenges that lied within it. In building Komo, we gained great experience for our future interactions with Amazon's AWS.
What's next for Komo
- Improve and expand technology to be able to identify more colors and clothing types
- Rank and sort results with user input
- Design and create a user friendly UI
Log in or sign up for Devpost to join the conversation.