Eleftherios Ioannou

I'm a PhD researcher at the  Computer Vision Research Group of the   University of Sheffield under the supervision of  Dr Steve Maddock.
My research 's underlying topic is Style Transfer which can be defined as the process of transferring the style of one image onto an input content image.

eioannou06 [at] gmail.com

Scholar  /  Github  /  Medium

Research

I'm interested in neural style transfer, computer vision, computer graphics, machine learning, image processing, computer games and augmented/virtual reality. During my PhD, I have explored the utilization of depth and other data for the improvement of the quality of artistic style transfer models. I am also intrigued by the intersection of computer science and art and I have studied computational aesthetics and computational creativity. I also have a deep interest in generative techniques and especially methods that can be utilized for synthesizing art (e.g., diffusion, GANs).

We introduce a new, highly controlled synthetic dataset that encompasses the essential attributes required for analyzing OoD behavior.

We use a perceptual quality-guided knowledge distillation framework and train a compressed model inspired by work in image quality assessment of 3D renderings, which substantially reduces both memory usage and processing time with limited impact on stylisation quality.

We provide an in-depth analysis of existing evaluation techniques, identify the inconsistencies and limitations of current evaluation methods, and give recommendations for standardised evaluation practices.

Utilizing G-buffer data enables the stylization process to be more aware of the geometric and semantic aspects of a game scene. G-buffer information utilized during inference time improves the stability of the stylizations, and offers a controllable way to stylize computer games.

We inject depth-aware NST as part of the 3D rendering pipeline.

A depth encoder network encodes ground-truth depth information which is fused into the stylization network. We employ ConvLSTM layers in the encoder, and a design a loss function based on calculated depth.

Empower teacher education using a novel Virtual Reality (VR) based approach. Development made in Unity.

Our approach uses a deep residual convolutional network with instance normalization layers that utilizes an advanced depth prediction network to integrate depth preservation as an additional loss function to content and style.

We support the systematic exploration of landscape archaeology through time, from prehistory to today, through the design and development of an Augmented Reality (AR) application.

We present an AR art app, running in real time on a smartphone, that can be used to bring to life inanimate objects such as statues.

The project produced Augmented Reality (AR) software that allows the users to view the stones standing and interact with them in situ.

Miscellanea

• cloud.torrent installation at the BufferFringe 2023
• Senrosial Entrapment Artistic installation at the Youth Makerspace

• On Tokens
• Neural Style Transfer: Past, present, future
• Semantic Segmentation of Water Bodies in Images
• Football 3.0: The beautiful game in the age of Artificial Intelligence
• Diffusion: The Physics principle behind the success of Modern AI Art

Code stolen from jonbarron/jonbarron_website.