Actions and results

Actions apply to the checked images, not simply the image open in the QuPath viewer. Confirm the image selection, assigned module and Scope before starting.

Fill, Redo and Metrics

Action

Use it to

Fill

Run missing model output while keeping existing output where it can be reused.

Redo

Run models again and replace their previous output.

Metrics

Recalculate measurements from the current objects without rerunning models or processing scripts.

The model selector defaults to All models. To choose individual models, the checked images must share the same installed module. This selector controls model steps, not script steps: processing scripts follow the recipe’s order. Fill and Redo also calculate measurements when the module defines them.

A module without measurements can still produce objects, but Metrics and HistoForge’s Excel Export are unavailable for it.

Set Module assigns the module used for subsequent actions; it does not remove existing objects. Measurements may include objects from other modules or manual annotations when their classes match the measurement script. Use a separate test project when comparing modules so their results are not counted together.

Analysis scope

Scope

Area analyzed on each checked image

HistoForge ROIs

Rectangles created with the HistoForge ROI tool.

Annotations by class

All eligible QuPath area annotations with the selected class.

Whole slide

The image, without requiring ROIs.

Follow Your first analysis for either ROI method, or Whole-slide analysis for an entire image.

Switching between ROI and whole-slide output for the same model requires Redo. Read its confirmation before accepting a replacement.

Using existing annotations

Annotations by class includes all eligible annotations with the chosen class on every checked image. Highlighting an annotation does not restrict this selection. Class names must match exactly: Tissue does not include Tissue: Subclass. If an image has no matching area annotations, HistoForge reports the problem; it does not analyze the whole slide instead.

Outlines may be irregular, contain holes or have several disconnected parts. They must stay inside the image and must not overlap, although their edges may touch. Points, lines, detections and annotations produced by HistoForge are not used as analysis boundaries. The ROI tool’s size restriction does not apply to these outlines; the image and model limits below still apply.

Both locked and unlocked annotations can be selected. Fill and Redo lock and save the boundaries before analysis, and model predictions are clipped to those outlines. To change a boundary later, use QuPath’s Unlock, edit it, save with Save Project & Refresh, then use Redo to replace its results. Metrics and Export do not change annotation locks.

Inference precision

BF16 can run model predictions faster on supported cloud GPUs. Shorter GPU runs can also reduce compute costs. Speed and cost savings depend on the model, image and GPU. BF16 uses fewer digits in some calculations, so predictions and measurements can differ from Default.

Open Settings → Inference precision to choose the numerical precision used for new Fill and Redo runs.

  • BF16 (if applicable) is selected initially. It uses BF16 when the model and GPU support it; otherwise it uses Default.

  • Default uses the model’s standard numerical precision.

Changing this setting does not alter existing results, unfinished runs or training. Use Redo to replace existing results with the selected precision.

Saving results

HistoForge may save results from several regions together. A region finishing processing does not mean its results have been saved. Stopping the action or closing QuPath before the save can lose unsaved results; results already saved to the project are kept.

An action can report Partial when only some results were saved or measurements failed. Check the image and Jobs → Details before continuing. For interrupted work, follow Resume unfinished work.

Image preparation limits

HistoForge checks the image source and requested region against the limits of the extension, connected backend and selected models. Opening a slide in QuPath or analyzing a small ROI does not establish that the same model can analyze the whole slide.

For prepared images, size estimates include all RGB pixels in the required bounding rectangle plus encoding overhead, including any context needed by the models. This is not the polygon’s area or the compressed slide size. An irregular region may therefore require more image data than its tissue area suggests.

HistoForge automatically chooses between preparing image data from QuPath and uploading an eligible local slide with its required companion files. Original-slide upload can use OpenSlide or Bio-Formats, following the image and reader selected in QuPath. Small regions may still use prepared images. See local image guidance for import steps and tested examples.

Each prepared file is limited to 16 GiB, or the backend’s limit if lower. The standard backend accepts an original-slide upload of up to 32 GiB in total, including packaged companion files. These are upload limits, not maximum slide sizes or guarantees of model support. Larger inputs may use several prepared images, each within its limit. Training has separate limits.

An original upload includes the entire slide and its metadata, even for one ROI. Read Data and privacy before uploading. Slides opened directly from S3 use prepared image data, not original-file uploads.

For a size error, follow Image-size limits. Do not change a model’s resolution just to bypass a limit: that changes the scale of tissue it sees.

Manual corrections

Model results are ordinary QuPath objects. After editing their shapes or classes, save the image and run Metrics to update the measurements.

Warning

Redo replaces output belonging to the selected models, including objects you have manually edited. It is not an undo operation. Manually created objects that do not belong to those models are preserved.

A model that successfully finds no objects can replace previous output with an empty result. This is different from a failed model run.

Reading the image table

Column

Meaning

Status

Whether output from the assigned module is present in the selected scope. It does not confirm that every model step ran or that the output is accurate.

Last Run

The outcome of the latest action in this QuPath session, such as Saved, Partial or Failed.

QuPath stores the analysis regions and result objects in the image data. Displayed HistoForge measurements are temporary and can be recomputed; export a workbook to retain them. The Last Run message is also session-specific.

If an action fails, read its complete message before repeating it. See Troubleshooting.