Computer Vision and Machine learning projects
An optimization and computer graphics project that involved the use of the computer graphics library ODAK to ray trace a simulation of looking through an AR optical component made up of pinholes. The component is then optimized with PyTorch by stochastic gradient descent. The result was a 96% reduction in error between the initially rendered images and the optimized images.
- Created a lane detection model for autonomous vehicle
- Model uses MatLab image processing libraries for edge detection and image processing
- Repositions vehicle based on the lane markings to centre it
- Model Tested on images and videos
Poisson image editing is a technique used to seamlessly blend together 2 images without artefacts. It is a type of digital image processing that operates on the differences between neighboring pixels, rather than on the pixel values directly. Poisson blending aims to copy the gradients of the source image into the destination image.
DCGANs are an updated version of generative adversarial networks that include the use of deep convolutional layers. Using this technique, I created a model that generates new images of popular show characters known as Pokémon based on a dataset on Kaggle containing images of existing characters. Code
This project uses convolutional neural networks(CNN) and a Kaggle dataset with images of faces with various emotions to create a web app capable of classifying the user's feelings. The web app is hosted via Streamlit and the user can upload an image of their face and receive a prediction of a feeling based on that image. Below are some images from the dataset used for testing with the predicted (pred) and actual labels (label) shown
A generative AI project that uses an unsupervised deep learning algorithm known as the generative Adversarial network. The popular computer vision dataset MNIST fashion dataset was used to train the model and the model generated images that were distinguishable from MNIST images. code