Segmentation errors often reflect gaps in the training data.

Review the ground truth, add missing examples, retrain, and iterate.

Custom model training gives you direct control over model behavior.

Guidelines for ground truth annotation: https://docs.webknossos.org/webknossos/tutorials/trainingdata_annotation.html

🌟Enhancement🌟

Open datasets or jump straight to annotations - right from the command palette.

Press Ctrl/Cmd + P from anywhere in the UI, then scroll or start typing to find what you need.

“And from a facility perspective, everything is automated. No one has to manually convert, upload, organize, or assign permissions.”

At EMBL P14, reconstructed tomography datasets are automatically converted and transferred to WEBKNOSSOS through the API.

Read more: https://medium.com/webknossos/from-beamline-to-browser-automating-high-throughput-x-ray-imaging-with-webknossos-at-embl-p14-dc992159574f

Run automated alignment in WEBKNOSSOS and improve the results with manual matches when needed.

Maintain control over challenging areas while keeping the workflow efficient.

https://www.youtube.com/watch?v=8IEl4m_Zg1A&list=PLpizOgyiA4kH-MbmYUpLcYf9RV8zuhDJ9&index=3

WEBKNOSSOS AI Grant 🔬

We are offering compute credits to support selected research projects applying AI segmentation to volume EM datasets.

Interested? Send us a link to your dataset in WEBKNOSSOS and a short description of your analysis goals at [email protected].

Train or refine custom segmentation models directly in WEBKNOSSOS.

Annotate ground truth, train the model, test on a small region, iterate, and run on the full dataset.

Dataset showcase ✨

In this study, Chen et al. (ScienceAdvances, 2025) investigate the ultrastructure and function of lamellar Schwann cells in Pacinian corpuscles, revealing their active role in mechanosensory processing. Explore the segmented dataset in WEBKNOSSOS: https://wklink.org/3248

At the EMBL P14 beamline, tomography datasets are automatically transferred into WEBKNOSSOS minutes after reconstruction.

Angelika Svetlove explains how the team built this pipeline - and why analysis, not acquisition, is now the main bottleneck.

Read the interview: https://medium.com/webknossos/from-beamline-to-browser-automating-high-throughput-x-ray-imaging-with-webknossos-at-embl-p14-dc992159574f

Run pre-trained segmentation models for SEM data in WEBKNOSSOS.

Test on small bounding boxes, inspect merge and split errors, and run on the full dataset once validated.

View the tutorial: https://youtu.be/lZS2tOJYErw