Mert Kıray

Computer Vision & Deep Learning Researcher

About Me

I am Mert Kiray, a dynamic figure in the realms of computer vision, deep learning, and software engineering. I hold a Bachelor of Science in Computer Engineering from Koç University, Istanbul, and a Master of Science in Informatics from the Technical University of Munich, where I distinguished myself academically

Professionally, I have taken on significant roles such as a Computer Vision Student Researcher at TUM, delving into contrastive learning methods using point clouds. As a Co-Founder and CTO at Octovan, I led teams and spearheaded technology decisions, creating a resilient API backend with Spring Boot and Hibernate. In another entrepreneurial venture as Co-Founder and AI Researcher at Jigglypop, I was at the forefront of predicting customer actions through deep learning techniques.

My master’s thesis and various projects reflect my expertise in integrating multi-modality context and self-supervised learning to advance 3D computer vision technologies with applications from autonomous vehicles to medical diagnostics.

With my proficiency in languages like Python, Java, and C/C++, and in using frameworks such as PyTorch, NumPy, and AWS, I have positioned myself as a thought leader and innovator at the intersection of computer vision and deep learning. My contributions have furthered academic research and led to tangible technological advancements in multiple sectors.

Interests

Computer Vision
Deep Learning
Machine Learning
Software Engineering
Cloud Computing

Education

M.Sc. in Computer Science, 2023
Technical University of Munich
B.Sc. in Computer Engineering, 2018
Koç University

Experience

Computer Vision Student Researcher

Technical University of Munich

April 2022 – April 2023 Munich

Worked as a research assistant in the 6G digital twin project at the Chair of Media Technology at TUM.
Conducted research on different contrastive learning methodologies on point clouds for different downstream tasks.

Software Developer | CTO | Co-Founder

Octovan

June 2017 – August 2020 Istanbul

As CTO, orchestrated developer teams and technology selection.
Created a resilient API backend from scratch using Spring Boot and Hibernate.
Engineered AWS-based infrastructure with EC2, S3, ECS, and Lambda services.

AI Researcher & Co-Founder

Jigglypop

September 2018 – September 2020 Istanbul

Created a customer action prediction using seq2seq approach in Python with PyTorch.

Teaching Assistant

Koç University

July 2017 – October 2017 Istanbul

Helped students to solve their questions about mobile device programming codes and graded projects.

Software Developer Intern

Accenture Digital

August 2016 – November 2016 Istanbul

Developed machine learning packages for a robotic arm.

Projects

Self-supervised Learning on EEG Data for Emotion Prediction

This project involves using self-supervised learning on EEG data to predict emotions, addressing the challenge of expensive and error-prone labeled EEG data by leveraging the unlabeled TUH dataset for pretraining.

Dream Challenge: Predicting gene expression using millions of random promoter sequences

The project aims to predict expression values from DNA sequences using an end-to-end Graph Attention Network, a sophisticated machine learning model designed to capture the complex relationships inherent in genetic data.

Laparoscopic Cholecystectomy

The project addresses the challenges in laparoscopic cholecystectomy, where surgeons face reduced depth perception from 2D imaging and visibility issues due to infection, obesity, and complex anatomy, increasing the risk of injury to surrounding tissues like the common bile duct and liver.

6G Digital Twin: Handover Prediction

This project aims to develop a 6G Digital Twin for handover prediction by constructing a detailed digital replica of a scene using LIDAR scans, and employing sophisticated scene understanding methods to enhance the accuracy and efficiency of digital twins in 6G networks.

End-to-end Holistic 3D Scene Understanding with Attention

Developed an advanced system for holistic 3D scene understanding from a single RGB image, using attention mechanisms to accurately predict object shapes, poses, and room layout.

3D Object Part Segmentation with Self-supervised Learning

Validated the effectiveness of self-supervised learning in achieving state-of-the-art results with limited label availability, revolutionizing the approach to 3D object part segmentation.

Weakly-supervised Semantic Segmentation through Projective Cycle-consistency

Employed self-supervised segmentation techniques to autonomously learn intricate scene understanding tasks with limited annotations, specifically in medical contexts such as surgical operating rooms.

Lidar Constraint NeRF on Outdoor Scenes

Utilized Lidar as an additional depth constraint for Neural Radiance Fields (NeRF) on outdoor scenes, expanding the field of view and improving scene understanding in autonomous driving applications.

Publications

Mert Kıray (2023). Exploiting Multi-Modality Context for Enhanced Online Adaptive Pseudo-Labeling of Point Clouds.

PDF Cite Dataset