Okan Köpüklü
Senior Researcher at Microsoft
Munich, Bavaria, Germany
Summary
👤
Senior🎓
Top SchoolOkan Köpüklü is a senior researcher and PhD candidate in Electrical and Computer Engineering at TUM with eight years of experience specializing in computer vision, pattern recognition, and deep learning. Now at Microsoft, he focuses on real-time spatiotemporal action and gesture recognition, bridging academic rigor with production-ready ML engineering. His open-source contributions include practical enhancements to prominent 3D CNN and real-time gesture repositories—adding ResNet/ResNeXt backbones, fixing FLOPs utilities, and adapting pipelines for multiple datasets—demonstrating an emphasis on reproducible, efficient models. He has a strong systems background from industry stints at Intel and BMW, which informs his approach to deployable research and dataset engineering. Colleagues value him for turning complex action-localization research into robust code and real-time demonstrators.
8 years of coding experience
3 years of employment as a software developer
Doctor of Philosophy - PhD, Electrical and Computer Engineering, Doctor of Philosophy - PhD, Electrical and Computer Engineering at Technische Universität MünchenBachelor of Science (B.Sc.), Electrical and Electronics Engineering (3.7 / 4.0), Bachelor of Science (B.Sc.), Electrical and Electronics Engineering (3.7 / 4.0) at Boğaziçi Üniversites,
Turkish, İngilizce, German
Github Skills (18)
pytorch10
python10
machine-learning10
mask-rcnn10
resnet10
computer-vision10
faster-rcnn10
video-processing9
deep-neural-networks9
shell8
sh8
modeling8
model-optimization8
trainings8
script8
Programming languages (4)
Github contributions (5)
okankop/Efficient-3DCNNs
Apr 2019 - Oct 2021
PyTorch Implementation of "Resource Efficient 3D Convolutional Neural Networks", codes and pretrained models.
Role in this project:
ML Engineer
ML EngineerContributions:51 commits, 1 PR, 44 pushes in 2 years 6 months
Contributions summary:Okan made several modifications to the project, including the addition of ResNet and ResNeXt models. The changes included updating existing model files (SqueezeNet, MobileNet, C3D) and fixing a bug in the FLOPs calculation utility. Further contributions involved modifications to the main script, dataset loading, and a shell script. The user demonstrates work within the domain of 3D CNN models.
pytorchresnetresource-efficientefficientnetdeep-learning
wei-tim/YOWO
Nov 2019 - May 2021
You Only Watch Once: A Unified CNN Architecture for Real-Time Spatiotemporal Action Localization
Role in this project:ML Engineer
ML EngineerContributions:31 commits, 36 pushes, 1 branch in 1 year 6 months
Contributions summary:Okan primarily contributed to the model and training pipeline within the "yowo" repository. Their work included modifying the `model.py` file to update the backbone architectures for the 3D CNN, specifically resnet and mobilenet variants. They also modified training scripts and evaluation scripts. Furthermore, the user adjusted the dataset paths and parameter settings within configuration files to align with the jhmdb-21 and ucf101 datasets.
deep-learningcnn-architectureaction-localizationspatiotemporallocalization
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.