Test and share your module
Test in a separate QuPath project before using a changed module on study images. A successful installation checks the package; a successful analysis shows that it runs. Neither establishes the model’s scientific accuracy.
Test an update
Save the changed files in your Hugging Face repository and note the commit. A commit identifies that saved version of the files.
In Modules → Install module, enter
owner/name@commit, replacingcommitwith the full 40-character commit shown on Hugging Face. Reinstalling the same repository replaces its installed version after validation; there is no separate Replace action.Select the module and check Installed revision. Hover over it to see the full commit.
Check Workflow and View recipe.json against your changes before running them.
If the model, its settings, or the processing recipe changed, run Redo with All models on your test regions. Fill may reuse existing model output. Redo replaces those models’ previous output, including manual corrections to it, so do this in the test project.
If only metrics.groovy changed, use Metrics. It recalculates from current objects without
rerunning models or processing scripts.
Keep the prior working commit so you can reinstall it if needed. Installing a different version does not undo changes already saved in a QuPath project.
Check expected behavior
Test |
What to check |
|---|---|
Small representative region |
Correct class labels, plausible outlines, and measurements you can check independently. |
No matching objects |
Zero counts and empty distributions, not invented values or errors. |
Model within structures |
Accepted structures provide child inputs. With Redo, no accepted structures means no child analysis; also check the Fill behavior. |
Processing script run twice |
Class changes remain consistent; objects do not accumulate repeated changes. |
Measurements and export |
Units, individual values and grouped summaries agree with a hand-calculated example. |
Invalid file or script |
A clear failure, not a plausible-looking replacement result. Use only your test copy. |
Test class filters on a fresh image without previous results. A measurement can otherwise count eligible objects from another module that share the same class.
For scientific evaluation, use images independent of training and model selection. Describe the images, annotations, evaluation procedure and limitations; do not substitute a successful software test for that evidence.
Explain what you are sharing
The repository’s README.md is its model card. Include:
Purpose and inputs: task, tissue, stain, species, image resolution and supported use.
Outputs: classes, processing steps and measurements with units.
Evidence and limitations: training sources, evaluation images and methods, results, and known failure cases. Explain how example images were selected.
Sources and permissions: original models, datasets and scripts, citations, applicable terms, and required license files or notices.
Use and changes: installation instructions, the version tested, and what changed from the source or previous version.
Keep a copied model’s history, but revise claims that no longer describe your result. In particular, the source model’s evaluation does not validate a new fine-tuned model. See Hugging Face’s model-card guide for formatting.
Share the tested version
If you or your AI assistant saved the module directly on Hugging Face, you do not need to publish it again from QuPath. After testing and reviewing its contents and sharing terms, change the repository’s visibility on Hugging Face if you want to make it public. Do not publish an older installed copy over your newer edits.
To upload a module installed in QuPath, follow Publish a module.
That guide also explains visibility, license notices and the histoforge discovery tag.
You can publish under your own account or lab organization; contacting HistoForge is not
a prerequisite.
Finally, verify that the intended audience can open the repository and install the tested revision. To discuss contributing a dataset or model, email histoforge.contact@gmail.com.