Conference Agenda

Brief Description and Outline:

Helmholtz Imaging is developing tools to support the chain of AI supported image data analysis. For successful training and application of AI methods is the quality of the data and the ease of use of the trained models. In this workshop we will present two recent developments to support this. In the first hour we will introduce PixelPatrol, a novel development by the Helmholtz Imaging Team at the MDC. And the second part will introduce the Helmholtz Model Zoo, a platform to share and deploy your trained models to be used from the browser for inference.

PixelPatrol is an early-version quality control and data exploration tool for scientific image datasets, acting as an essential first step for you to understand your dataset before computationally intensive analysis. In a first step it systematically profiles your imaging data. And in a second stage interactive dashboard reports with statistics and visualizations are generated to ensure data integrity and quality.

This hands-on workshop introduces participants to systematic data curation practices for scientific image datasets using PixelPatrol, an open-source tool for dataset exploration and quality control.

Participants will learn to identify data quality issues before computationally expensive analysis, explore dataset characteristics interactively, and implement quality control workflows that prevent common pitfalls in imaging pipelines and AI model training.

The Helmholtz Model Zoo (HMZ) is a cloud-based platform enabling seamless sharing and inference of deep learning models across the Helmholtz Association. By automating model deployment and providing both web and programmatic interfaces, the HMZ lowers technical barriers to AI adoption in scientific research. Integrated with Helmholtz infrastructure (Helmholtz ID authentication, dCache storage, DESY’s HPC cluster with NVIDIA L40S GPUs), the platform ensures secure, scalable inference while maintaining data sovereignty. NVIDIA Triton Inference Server and Slurm manage GPU resources efficiently, supporting data-sets from gigabytes to terabytes. Virtual organizations enable fine-grained access control for specialized models across various scientific domains. 

We will also provide an outlook on how both developments will be integrated in the near future. With PixelPatrol it will be possible to generate fingerprints of the training data, which can be used in the context of the HMZ to support users to identify trained models where the fingerprint of the training data suggests successful inference to the users data.

-

Workshop Outline (2 hours):

PixelPatrol Overview

Tool architecture and capabilities

Installation and setup (live demonstration)

-

Workflows

-

Hands-On Session 1: Basic Quality Control

• Participants process their own datasets or a provided example - aqQua foundation model training data.

• Generating comprehensive quality reports

• Interpreting visualizations: file stats, metadata consistency, image statistics

• Interactive filtering and grouping strategies

• Identifying outliers, artifacts, and acquisition inconsistencies

-

Hands-On Session 2: Make pixelPatrol your own 

• Development of PixelPatrol packages and plugins

• Helmholtz Model Zoo Overview

• overview of the HMZ platform and its capabilities

• How to use a deployed model

-

Hands-On Session 3: Deploy your models to the model zoo

• Prepare your models to be deployed in the HMZ

• Overview on templates for the deployment

-

Summary & Outlook

-

Goals:

Pixel Patrol: This hands-on workshop introduces participants to systematic data curation practices for scientific image datasets using PixelPatrol, an open-source tool for dataset exploration and quality control.

Participants will learn to identify data quality issues before computationally expensive analysis, explore dataset characteristics interactively, and implement quality control workflows that prevent common pitfalls in imaging pipelines and AI model training.

With PixelPatrol it will be possible to generate fingerprints of the training data, which can be used in the context of the HMZ to support users to identify trained models where the fingerprint of the training data suggests successful inference to the users data.

We will also provide an outlook on how both Pixel Patrol and the Model Zool will be integrated in the near future.

--

Presenters Experience:

Ella Bahry is a Research Software Scientist in the Support and Engineering group at Helmholtz Imaging, MDC Berlin. Working across diverse scientific imaging modalities and datasets, she identified a critical gap in data quality practices that inspired the development of Pixel Patrol. Ella is passionate about addressing this real-world need and enjoys thinking about and teaching researchers of best practices.

Deborah Schmidt is Group Leader of the Support and Engineering team at Helmholtz Imaging, MDC Berlin. With expertise in bioimage analysis, visualization, and scientific software development, she leads initiatives to create open-source, community-driven tools for imaging researchers. Deborah has extensive teaching experience in workshops and tutorials and is deeply committed to making robust data practices accessible across the imaging community.

Ella and Deborah are the co-lead of PixelPatrol development and both teach courses at HIDA.

Hans Werners is a Research Software Scientist in the Support and Engineering group at Helmholtz Imaging, DESY Hamburg. Hans is the key developer of the Helmholtz Model Zoo. He brings hands-on expertise from both developing the platform and actively supporting researchers in deploying their models

Philipp Heuser is head of the Helmholtz Imaging Support and Engineering team at DESY and deeply involved in the development of the HMZ.

-

Target Audience:

This workshop is designed for:

• Imaging researchers working with large or heterogeneous imaging datasets

• Scientists preparing datasets for AI/deep learning workflows

• Data curators and research software engineers supporting imaging facilities

• Researchers contributing to or using large-scale imaging datasets

• AI experts, machine learning engineers, and researchers who have experience training or working with deep learning models and want to make their models accessible to the broader Helmholtz community

Prerequisites:

• Basic familiarity with scientific image data. No programming expertise required—the tool provides both browser-based UI and command-line interfaces. 

• Participants should bring a laptop and (optionally) a sample dataset they'd like to explore.

• Participants who want to deploy their model can get in touch with us beforehand as well under support@helmholtz-imaging.de

• Participants should bring laptops and ideally datasets.

-

Keywords:

data curation, image quality control, dataset exploration, deep learning workflows, data integrity, metadata validation, FAIR research, model deployment, Helmholtz Model Zoo, inference