Computer Vision + Predictive Modeling
During the second half of the Hack Reactor program, I had the opportunity to work on a contract project for PIXLEE. PIXLEE is an early stage Andreessen-Horowitz backed start-up that builds photo dashboards for clients such as Coke, the 49ers and Yamaha.
After talking to the team, I decided to focus on two key problems; being able to filter images that contain a certain template, and developing a predictive model to evaluate the virality of a photo.
The most challenging component of the computer vision project was researching the appropriate technologies to use, and setting up an environment in which I could use these technologies. I ended up using an Ubuntu EC2 instance, and using OpenCV/SimpleCV as the core computer vision module. Code for this project is available at https://github.com/sheltowt/PIXLEE-public. A visualization of the resulting data was made by Greg Palmer and is viewable at http://computer-vision-example-pixlee.nodejitsu.com/.
The predictive modeling project yielded results similar to what I expected. The number of followers a user had, the time of day and the day of the week in which the post was created, and the hashtags the poster used, all had significant value in predicting which photos were most likely to recieve the highest number of likes. The code for this project is available at https://github.com/sheltowt/PIXLEE_predictive_modeling.
