Tero Karras
Research Scientist at NVIDIA
Helsinki, Mainland Finland
Summary
🤩
RockstarTero Karras is a research scientist and principal ML engineer based in Helsinki with eight years of experience specializing in generative adversarial networks and production-ready model tooling. He has been a key contributor to landmark NVIDIA projects such as StyleGAN, StyleGAN2 and progressive growing of GANs, implementing core network code and maintaining critical infrastructure like pretrained model distribution and training configs. His work blends deep research rigor with practical engineering—updating dependencies, improving robustness across TensorFlow/NCCL stacks, and migrating checkpoints to scalable storage (AWS) to support broader reproducibility. Colleagues benefit from his ability to translate academic papers into clean, usable TensorFlow implementations and to harden training pipelines for real-world use. An often-overlooked strength is his attention to ecosystem details—package checks, dataset utilities, and documentation—that make cutting-edge models accessible to practitioners.
8 years of coding experience
Github Skills (13)
computer-vision10
machine-learning10
deep-learning10
tensorflow10
python10
generative-adversarial-network10
pre-trained-model10
neural-network9
debug9
hdf9
image-generation8
stylegan8
pytorch3
Programming languages (1)
Github contributions (5)
NVlabs/stylegan2
Dec 2019 - Oct 2021
StyleGAN2 - Official TensorFlow Implementation
Role in this project:
ML Engineer
ML EngineerContributions:17 commits, 16 pushes, 1 branch in 1 year 10 months
Contributions summary:Tero's contributions primarily focus on updating the project's dependencies on pre-trained networks and model checkpoints, switching download locations from Google Drive to Amazon AWS. These changes involved modifying various Python scripts that handle the retrieval of pre-trained models and classifiers used within the StyleGAN2 framework. Furthermore, they updated the documentation with links to the official StyleGAN2-ADA, StyleGAN3, and Alias-Free GAN projects.
stylegancycleganimage-synthesisdeep-learningstylegan2
NVlabs/stylegan
Feb 2019 - Feb 2021
StyleGAN - Official TensorFlow Implementation
Role in this project:ML Engineer
ML EngineerContributions:14 commits, 11 pushes, 1 branch in 2 years
Contributions summary:Tero primarily contributed to bug fixes and improvements within the StyleGAN project, focusing on code related to running metrics and training configurations. They addressed issues related to TensorFlow, including fixing the NCCL package location and handling HTTP and Google Drive errors. Furthermore, they modified training configurations, including learning rate progressions and added configs for different resolutions of the FFHQ dataset.
stylegancyclegandeep-learninggenerative-adversarial-networktensorflow
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.