Inspiration
So you bought that couch. Now which coffee table will go with it? How about some accent chairs and wall art? Can you pick the right combinations? What would it take to turn everyone into a decorator?
What it does
You can take picture of things in your home and it will recommend what else that you can buy from Wayfair that will go along with it
How we built it
We do it in two steps. First step is powered by a deep neural network (trained over the weekend). It identifies closest matching product from Wayfair with the image you snapped. In the second step we recommend items from other related classes that we have compiled based on customer buying patterns.
Challenges we ran into
- Training a deep neural network over the weekend. It typically takes days and multiple iterations before it could actually product useful results. We did get decent results in the end but it could be much better with more training.
- Finding a capable hardware to run the complex network which we wanted to use.
- Making a reasonably responsive search service that could go over thousands of product images to find a match
- Making it all work with Wayfair mobile app
Accomplishments that we're proud of
We did overcome above challenges!
What I learned
Making deep learning work is very hard but not impossible. Our products are visually shopped. Its a huge opportunity for Wayfair if we can do this right.
What's next for DecorNet
Currently we recommend items in living room based on customer snapped pictures of sofas. In future we would train our model on more items, train it on larger training set and implement a snappy search to make it work in "real time". We can also let customer pick the best match from the search results.
Log in or sign up for Devpost to join the conversation.