DeepEM Playground:

Bringing Deep Learning to Electron Microscopy Labs

Hannah Kniesel1, Tristan Payer1, Poonam Poonam1, Tim Bergner2, Pedro Hermosilla3 Timo Ropinski1,
1Visual Computing Group, Ulm University 2Central Facility for Electron Microscopy, Ulm University 3Computer Vision Lab, TU Vienna

Image to Value(s)

Object Counting in Electron Microscopy

Image to Value Get Started 🚀

Image to Image

Semantic Segmentation of Cellular Structures

Image to Image Get Started 🚀

2D to 3D

Tomographic Reconstruction

2D to 3D Get Started 🚀
Paper

DeepEM Playground is a user-friendly, interdisciplinary tool designed by deep learning (DL) experts to help electron microscopy (EM) labs leverage DL without requiring expertise in coding or AI. It simplifies model training, testing, and deployment, making DL accessible to researchers at any experience level, guiding them towards deeper understanding of AI.

Key Features

  • ✅ Easy Setup – Thanks to the use of Lightning AI, users can train models without configuring GPU environments.
  • ✅ Seamless Collaboration – Easily share model training procedures and model weights with colleagues.
  • ✅ Structured Use Cases – A clear structure of EM specific use cases provides an overview of possibilities to bring DL into EM labs.
  • ✅ Standardized Workflow – Standardizing the implementation of use cases provides a clear structure and simplifies the learning curve.

Workflow

Each use case implementation follows a standardized workflow, structured into:

  • 🔹 Development: Users train, fine-tune and evaluate models, gaining hands-on experience.
  • 🔹 Inference: Trained models can be used immediately, including models shared by colleagues.

DeepEM Playground empowers EM researchers to harness DL effortlessly, fostering innovation and collaboration across labs. It provides an interface for DL experts to make their work available to EM researchers by contributing their work to the playground.

Learn More

Use Cases

Use cases are structures into three distinct tasks:

  • 📌 Image-to-Value(s): Extracting key numerical data from images (e.g. quantification of virus particles).
  • 📌 Image-to-Image: Transforming one image into another (e.g. segmentation).
  • 📌 2D-to-3D: Reconstructing 3D structures from 2D images (e.g. tomographic reconstruction).

Each task is demonstrated through use cases. A use case is developed by DL experts and made available through our playground. Each use case is defined by its primary focus (like segmentation) and its exemplary application (like segmentation of cellular structures). With a plug-and-play approach, researchers can easily adapt the application of the use case within the primary focus area (like segmentation of mitochondria) simply by changing the data—no coding required.

Learn More

Getting Started

Getting started is quick and easy! Follow these steps to explore our platform:

  • 🚀 Explore Use Cases: Browse available applications and find one that fits your needs.
  • 📂 Upload Your Data: Simply provide your own dataset to adapt an existing use case.
  • ⚙️ Run and Analyze: Apply deep learning models with no coding required and evaluate results effortlessly.

Clicking the button below will guide you through the fundamental concepts of the DeepEM Playground in a step-by-step manner.

Get Started

Your Contribution

The success of DeepEM Playground relies on its users.

We encourage EM experts to utilize the tools provided in this playground to learn about, integrate and adapt DL solutions into EM workflows, simplifying and enhancing image analysis.

We invite DL experts to contribute their methods and training code, making them accessible to the wider community and fostering interdisciplinary collaboration.

If you are interested in contributing your work, please click below:

Contribute Here

BibTeX

BibTex Code Here