In this document, I showcase several deep learning/machine learning projects that I completed during my time as a university student. These projects cover a variety of applications in natural language processing, computer vision, and generative modeling.
My BSc thesis. I located people in a single image using the tinyface detector (it was clearly the best of all methods we observed (at that time)) and then devel- oped an algorithm that created trajectories and followed/counted people from frame to frame in a video. Thanks to Nikola Banić for his guidance while i was working on this.
We had a dataset composed of 250k song lyrics and their respective genres. The part I worked on was creating a tfidf weighted fasttext vector from the top 10 ranked words in the song, and then finding the best classification algorithm. To no one’s surprise, it was XGBoost
We trained a pix2pix GAN. The dataset had neutral facial expression image, an image with some emotion, and the corresponding emotion vector, the results were surprisingly good. Network inputs were: face image with neutral expression + emotion encoded in a vector output: face image with drawn emotion
This was done as my graduate project, the dataset I had was the same one used in the Retouch Challenge. Experimented with different/new loss functions and observed how the end result changes with respect to the loss function with interesting results
A seminar on GANs and how they could be used for superresolution in images, and various difficulties the traditional approach had in achieving superresolution. I’m a big fan of how GANs work for superresolution.