This was the first project for my computational photography class with professor Alexi Efros. It involved aligning and enhancing the Prokudin-Gorskii photo collection.
The project was all about using the gradient domain. Ranging from edge detection convolution filters to multiresolution blending.
In this project, I create some face merging and averaging and then apply machine learning to automate the process.
The first part of this project used a CNN to classify the fasion mnist set. The Second used a U-Net style architecture to segment images of building dataset.
This project had 4 parts including automatic image captioning using first vanilla recurrent neural nets and then LSTM RNNs followed by a section on network activation visualization and finally a cool section on style transfer
This project deals with sequence processing. I use both RNNs and transformers to complete a variety of tasks. Note: private until May 15.
The final project for this class was an open problem in computer vision. The goal was to improve a CV system trained on tiny imagenet to be more resistant to natural adversarial examples as mentioned here.