In this project, we used PyTorch to implement the faster RCNN for detecting objects (Marine Life). The objective of this project is to help showcase the potential of Machine Learning for conservation. In our case, we trained on the Brackish Dataset which like many real-world observations is naturally noisy and low contrast. This is a regime in which AI can outperform human observers. Our implimentation utalized a RESNET 50 backbone CNN.
As we wanted to show the biodiversity monitoring implications we decided to keep track of the number of animals in each frame an preliminary example is shown in the video below.