Rock, Paper & Scissors Image Classifier:
This AI is based on a Convolutional Neural Network, utilising three 2D convolutional layers, two Linear layers, integrated Max Pooling and Dropout. This 98% accurate image classifier AI is trained on a large 18,500 image dataset, captured from split-up short videos of first-person hand shots. These poses of either rock, paper, or scissors were taken at various locations, light settings, and hand models. The AI processes these images and deduces their pose through 15 epochs of training. A full account on the implementation can be found in the detailed report.
Retro-futuristic Car Design Image Generator:
Alternatively, this AI is based on a Deep Convolutional Generative Adversarial Network (DCGAN) Neural Network (NN), utilising two NNs where one Generates the images, and the other acts as a Discriminator, evaluating the fake images against real images. As the NNs learn, the generator will aim to eventually make their results looks just as realistic as real results, thus producing photo-realistic results. The model is based on a generous dataset of 637 internet scraped classical car, modern car, and futuristic concept car images. This therefor required 340 epochs in batches of 50 images each to produce realistic looking AI produced car concept models. A full account of the implementation can be found in the detailed report.
Final Progress Videos