Çağrı Eser

Welcome! I am a Computer Vision and Machine Learning Engineer with over five years of experience in perception systems for autonomous vehicles. I completed my MSc in Computer Engineering at METU in 2024, where I also earned my BSc and a Minor in Mathematics. My thesis addressed class imbalance in long-tailed visual recognition, and my advisors were Sinan Kalkan and Emre Akbaş.

I am interested in computer vision, machine learning, and perception for autonomous vehicles. I like working on problems in long-tailed learning, multi-modal learning and geometric methods applied for robust and interpretable machine learning. My current research explores the behavior of deep learning models under various forms of imbalances.

News

June 2026 - “Evaluating Pretrained Music Embeddings for Cross-Performance Jazz Standard Recognition” has been accepted to ICML2026 @ Workshop on Machine Learning for Audio!
- More details to come soon!
March 2026 - I’m serving as a reviewer for SIU2026.
Our latest work, “Intrinsic Dimensionality as a Model-Free Measure of Class Imbalance”, is now available on arXiv.
- February 2026 - IDIM has been published in Neurocomputing!
I’m serving as an adjunct lecturer for CENG240 for the Spring 2024 semester, a course for Programming with Python for Engineers.
- Weekly course materials can be found here in the form of Jupyter notebooks.
I have completed my Master of Sciences degree of Computer Engineering at METU. I’d like to give special thanks to my supervisors Prof. Sinan Kalkan and Prof. Emre Akbaş for their continual guidance throughout my studies.
- My thesis was about mitigating the effects of class imbalance in long-tailed visual recognition using intrinsic dimensionality. It can be accessed on OpenMETU through this link.
- cGPA: 4.0/4

Publications and Preprints

Evaluating Pretrained Music Embeddings for Cross-Performance Jazz Standard Recognition
Çağrı Eser
Accepted to ICML2026 @ Workshop on Machine Learning for Audio
We evaluate the effectiveness of music embeddings from audio pretrained models on the challenging task of standard recognition from jazz performances, and suggest a lightweight contrastive adaptation for retrieval-based approaches.
[PDF, code and details soon!]
Intrinsic Dimensionality as a Model-Free Measure of Class Imbalance
Çağrı Eser, Zeynep Sonat Baltacı, Emre Akbaş, Sinan Kalkan
Neurocomputing 674 (2026) 132938
We propose an alternative perspective on imbalance in long-tailed datasets, focusing on the intrinsic dimensionalities of classes in image space rather than their cardinalities.
[PDF] | [Preprint] | [Code]
Mitigating class imbalance in long-tailed visual recognition
through the use of intrinsic dimensionality
Çağrı Eser
MSc thesis
I concentrate on model-based and data-based definitions of intrinsic dimensionality and their relation to model performance, choice of architecture, choice of estimator and imbalance severity.
[PDF]