Inspiration
When we utilized Clarifai's services, we knew that we would not have access to lots of images that we can input to make it more accurate, so we wanted to see if we could automatically train this service to take a few sample pictures from a user then run through a google search to gather lots of related images and further train the model.
What it does
It takes a small number of manually inputted data with a given search term, which then runs it through google images and gathers a large collection of images. Then it runs those images through the image visualization predictor to make decisions on the presence of the object represented by the search term. This results in a large set of data that probably has the object, so when it is added to the data set, it increases the precision of the model with minimum user input.
To demonstrate the effectiveness of this process, we developed an app where users can upload pictures of the rooms in their house to identify potential fire hazards which were determined by a small set of user input images that contained the fire hazards.
How we built it
We created a backend using Java and Spark for the API to host a backend that would run our optimization code for Clarifai which would be queried from an android application which was developed using Android Studio.
Challenges we ran into
No one on the team had large amounts of experience using machine intelligence so it took us a while to see use cases for the technology and how we could contribute to it.
Accomplishments that we're proud of
We all focused on technologies that we were not familiar with and got the opportunity to expand our understanding of how to develop these other technologies. We are also proud that we were able to optimize small amounts of user input to create strong and adaptive models.
What we learned
Some of the skills that we picked up were specific skills related to Clarifai, but we can extrapolate these skills to understand how to interact with other 3rd party APIs as well as how to utilize these APIs into other projects. We also learned skills related to developing Android apps and servers running Spark.
Log in or sign up for Devpost to join the conversation.