Inspiration
I was looking forward to create a filter which everyone can love and feel. So first of all I tried many snapchat filters and studied what I liked the best about the filters. The thing is that, The filters I liked was either aesthetically very pleasing or funny. So I tried to ideate my filter to be a aesthetic one or an funny one. I crated about 5-6 projects until I finally had the idea for my filter "Brownie". The inspiration for Brownie was derived from my childhoods favorite filter which was one of the few filters in the nokia's old mobile phone namely sepia. The sepia effect is a well-known tonal editing technique that adds a warmer tone that gives an image a vintage or archival quality.
How I built it
At first I tried to make LUT filters, but I realized that they are very simple and cant attract somebody that often and we can already find so many of them in snapchat itself. So I decided to take the game to the next level, Machine Learning. This was my first time into ML, so I was curious and not confident at the same time. I researched at learned about ML and how to make models like onnxx model to be implemented in lens studio. then I started ML and provided the machine with an abstract image and the model was generated. I tried many iterations using many images until I found the perfect and aesthetically pleasing one. I created and trained a neural model for style transfer. It's a computer vision method used for reproducing an image (so-called content image) with style features of another one (so-called, reference image). With neural style transfer, it becomes possible to imitate painting techniques of famous artists. We will train our neural network on the famous COCO dataset that contains images of different objects.
Challenges I ran into
The main challenge was the Machine Learning part. I had no prior experience in the Machine learning field, so was not so sure about doing it. but I have been inspired by what machine learning can do by some example in snapchat. A challenge I faced was that, it takes a tremendous amount of time to the machine to learn and it was driving me crazy. But whenever I was seeing the results of the waiting I realized that waiting can result in such great outcomes.
What's next for Brownie
I could look forward to improve the filter by adding some interactions to it and may be a grain texture to it, also in terms of monetizing, it can be done through advertising by Partner with brands and companies to showcase their products or services within the filter.
Thank you
Log in or sign up for Devpost to join the conversation.