Profile Picture

Emre Girgin

Ph.D. Student / Research Assistant

Education

Ph.D. in Aerospace Engineering

Embry-Riddle Aeronautical University [2025–Present]

M.Sc. in Computer Engineering

Bogazici University [2021–2024]

B.Sc. in Computer Engineering

Bogazici University [2016–2021]

Experience

Robotics Researcher

TUBITAK BILGEM [2021–2024]

Research Interests: Computer Vision • Deep Learning • State Estimation • Uncertainty Estimation

Publications

Learning Rock Pushability on Rough Planetary Terrain

International Conference on Robotics and Automation (ICRA) - Workshop on Field Robotics, 2025.

EdgeAI Drone for Autonomous Construction Site Demonstrator

International Conference on Robotics and Automation (ICRA) - Workshop on Future of Construction, 2025.

Detection and Quantification of Occlusion for 3D Human Pose and Shape Estimation

International Conference on Pattern Recognition (ICPR) - Workshop on FBE, 2024.

Bidirectional Human Interactive AI Framework for Social Robot Navigation

International Conference on Robotics and Automation (ICRA) - Workshop on RTSS, 2024.

A Novel Occlusion Index

IEEE Signal Processing and Communications Applications Conference (SIU), 2023.

Selected Projects

Rock Segmentation Point Cloud

Point Cloud Based Rock Segmentation (2025)

A ROS2 package to segment rocks and extract surface characteristics. (ICRA'25)

Occlusion Index

Occlusion Index (2024)

Occlusion quantification tool for 3D Human Pose and Shape (HPS) Estimation datasets. Master Thesis

LatentMix

Saliency Map Based Adversarial Attack (2022)

A scoring-based black-box attack written in PyTorch that forms adversarial images by optimizing the network's confidence level with guidance from a saliency map.

Blender SMPL

SMPL Rendering using Blender (2022)

Example pipeline to load and render SMPL models using the Blender API.

LatentMix

Adversarial Robustness by Latent Code Composition (2021)

A PyTorch-based pipeline to improve the robustness of image classification models. It composes the latent code of original and attacked samples during adversarial training via an AugMix-like algorithm. (BSc Senior Project)

Detectron2 BDD100K

Detectron2 for Lane Segmentation (2021)

Using the Detectron2 API's Faster R-CNN pipeline for lane segmentation on the Berkeley Deep Drive 100K dataset.

Tutorials & Documents

Blog

I occasionally write about topics in robotics and computer vision or simply life.

Travel Notes