Starting Work on a New Project

My next project will be focused on machine learning and neural networks. Using pyTorch, I will be able to train a convolutional neural network to recognize images, specifically recognizing familiar faces.

About the Project

I was able to get the first tests up and running with the basic image recognition example using the CIFAR10 Dataset where the model was able to recognize different objects such as cats, dogs, frogs, horses, and cars.

The CIFAR10 Dataset is extremely dense, as it contains thousands of images which makes it optimal for learning the basics of training neural networks.

Why This Matters

Learning the basics of convolutional neural networks will aid me in future game development projects.

For example, if I was to make a game or sequence inside a game similar to Google’s Quick, Draw! Website, knowing the basics of a convolutional neural network would help.

I was inspired by Sebastian Lauge’s 2022 Video on teaching a neural network to identify doodles in Unity, and I look forward to implementing something similar into my game engine.

The First Tests

The first few tests of the model returned mixed results, as it was able to recognize about 50% of the 10,000 images that were served to it. After a few hours of training, its accuracy rose ~20%.

I will be updating more about this project in the next few days, to follow closer, check out the repository