Robin Rombach
PhD Candidate
Baden-Württemberg, Germany
Summary
🤩
RockstarRobin Rombach is a PhD candidate and machine learning engineer in Baden-Württemberg, Germany, with seven years of experience building and shipping generative image models. He actively contributes to high-profile open-source projects from CompVis (latent-diffusion, taming-transformers), implementing practical improvements such as a VQGAN loss with codebook statistics, PyTorch Lightning 1.0 upgrades, and a WebDataModule for large-scale webdataset loading and inpainting masks. His work bridges research and engineering, translating model innovations into production-ready data pipelines and inference tooling for high-resolution image synthesis. As a long-term PhD student at LMU Munich, he couples rigorous academic focus with hands-on system-level fixes that accelerate training and usability.
7 years of coding experience
Github Skills (18)
transformers10
pytorch10
python10
diffusion-models10
machine-learning10
pytorch-lightning10
webdataset10
data-loading10
image-generation10
computer-vision10
image-preprocessing10
preprocess9
preprocessing9
load-data9
inpaint9
Programming languages (2)
Github contributions (5)
pesser/stable-diffusion
May 2022 - Aug 2022
Role in this project:
ML Engineer
ML EngineerContributions:83 commits, 45 pushes, 1 comment in 2 months
Contributions summary:Robin implemented a new data module for loading data from web datasets. This involved creating a `WebDataModuleFromConfig` class that utilizes `webdataset` for efficient data loading and preprocessing, including image transformations and batching. The user added support for image transformations, data filtering, and mask generation for inpainting tasks, indicating contributions focused on data loading and preparation for image-based models.
CompVis/latent-diffusion
Dec 2021 - Jul 2022
High-Resolution Image Synthesis with Latent Diffusion Models
Role in this project:ML Engineer
ML EngineerContributions:22 commits, 4 PRs, 11 pushes in 7 months
Contributions summary:Robin contributed to the implementation of a VQGAN loss function with codebook statistic evaluation, enhancing the model's training and performance. They modified the `ddpm.py` file, likely to incorporate these new loss functions and other model adjustments. Additional commits included adding new models and modifications to the `txt2img.py` script and other related files, indicating an active role in model development and inference processes. The user also worked on general model modifications and improvements related to inference.
pytorchimage-synthesisdeep-learningsynthesisresolution
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.