Experience

  1. Undergraduate Researcher

    Middle East Technical University
    • Worked with R. Gokberk Cinbis on weakly supervised semantic segmentation (WSSS) using the PASCAL VOC 2012 dataset, employing only image-level labels as weak supervision.
    • Analyzed the impact of confident region thresholds on AffinityNet performance, showing that threshold selection directly influences pseudo-label quality and overall WSSS pipeline effectiveness.
    • Devised novel evaluation criteria to optimize confident foreground and background thresholds, improving the quality of pseudo labels and addressing an important gap in existing research papers.
    Read Research Report
  2. Computer Vision / Machine Learning Intern

    Kuartis
    • Conducted a literature review on Weather Classification for Autonomous Driving. Collected data for various weather conditions, applied the CLAHE filter to images, and divided them into patches. Adopted model architectures for multi-frame input, trained, and tested several image classification models.
    • Performed a literature review on Semantic Occupancy Prediction. Presented a report summarizing state-of-the-art architectures, loss functions, and datasets.
    • Implemented a horizon detection pipeline for Marines using classical vision methods. Utilized the Canny edge detector to extract edges at different scales, fused the extracted edge maps, fitted horizon lines on the fused maps using the Hough Line Transform, and eliminated outliers with RANSAC.
    • Evaluated several trackers in the OpenCV library and determined that the CSRT Tracker is the most accurate while achieving real-time performance.
    • Delivered a presentation on dataset curation, neural architecture search (NAS), and hyperparameter optimization.
    Read Internship Report
  3. Research Intern

    ROMER Research Center
    • Worked on a project that aims to reposition a robot arm concerning the position, orientation, and alignment of a captured image of an object, ensuring it appears as though the arm never moved.
    • Researched, implemented, and compared several feature-based image-matching algorithms, including SIFT, SURF, FAST, ORB, and the SuperGlue model, for accurate image alignment.
    • Applied RANSAC and NN ratio matching algorithms to eliminate weak matches, enhancing the robustness of the image-matching process.
    Read Internship Report

Education

  1. B.Sc. in Computer Engineering

    Middle East Technical University
    • CGPA: 3.91/4.0

    Courses included:

    • CENG483: Introduction to Computer Vision
    • CENG501: Deep Learning
    • CENG796: Deep Generative Models
  2. Double Major in Mathematics

    Middle East Technical University
    • CGPA: 3.82/4.0

    Courses included:

    • MATH371: Differential Geometry
    • MATH358: Partial Differential Equations
    • MATH478: Mathematical Aspects of Cryptography