Optimize Animal Tranquilizer Aiming System

Important

What's interesting is that the training data is fully cats and dogs, but the model works with new animals too! The model is trained on the Oxford IIIT Pets dataset, which contains 37 classes of pets. The model can be used for any animal.

Screen.Recording.2024-12-20.at.10.30.50.mov

Term project for the course: Neural Networks and Deep Learning. This project emphasizes improving the tracking of animals in real-time to ensure that the tranquilizer accurately aims at the target.

The dataset used is Oxford IIIT Pets dataset. The dataset itself contains the respective maskings.

We have developed a model to track the motion of the animal by masking the animal to represent the trace of the tracking achieved by our model.

We have worked with three models, namely:

1. Simple Unet tensorflow trained in Google Colab TPU
2. Unet CNN tensorflow trained in Google Colab TPU
3. FPN with ResNet34 pytorch trained in Apple M2 Pro with Metal GPU 19 cores.

Evaluation Plots

Simple Unet

Unet CNN

FPN with ResNet34

Model Comparison

Inference

To showcase the results, we use images and videos to prove its effectiveness.

Unet CNN

FPN with ResNet34

Screen.Recording.2024-12-20.at.10.30.50.mov