Unlike the rest of my team, I am not pursuing a computer engineer and computer science major. Instead I am a pursuing a degree in wildlife ecology and conservation. As an intern, we spend 25% of our time watching videos so that we can write down the time in the video the animal being researched, usually a deer, appears. This is not only a waste of time on our end, but also a waste of money for the company.
What it does
So, to help with this issue, we decided to make a machine learning algorithm that takes a video and makes note of when a deer is on screen and when it leaves the screen.
How we built it
We used pytorch to teach our algorithm using 100 pictures of deer and 3000 pictures of other animals and empty landscapes. We used pug, express, nodejs, and semanticsui to build our front end interface. We used python for our back end interface.
Challenges we ran into
None of us had ever used python or done machine learning before so it was a lot of trial and error on our part. We tried using open cv first but found that it was not suitable for our needs so we had to switch to pytorch which wasted half of our hacking time.
Accomplishments that we're proud of
We were able to actually train pytorch to be over 90% accurate. We also did not kill each other during this hackathon even though we were all very stressed out and sleep deprived.
What we learned
We learned how to use python and also how machine learning actually works. We also tested out react, which we did not end up using, but it was good experience for us. We also learned that front end work is awful.
What's next for Fauna Finder
In the future we would like to expand this to other species so that we can reach all area's of wildlife research. This could improve the efficiency of our research and allow for our interns to do better things with their time.