Skip to content

Inference Tab

Overview

The Inference tab (InferencePage.qml, route Inference) is the cross-project Inference Monitor ("Monitor inference runs across all projects"). Rather than configuring new prediction parameters directly on this screen, new inference runs are launched directly from a slide or selected ROI inside the Viewer tab via Predict..., while this tab tracks every run across your projects in real time.

Inference page showing the cross-project Recent Inference Runs table with search bar, Status badges, and execution Details.

Availability & Prerequisites

  • Active Project Required: No — intentionally aggregates inference runs across all projects.
  • Prerequisite to Generate Runs: To submit a new prediction job, open a project in Projects, load a WSI in Viewer, select a model and target region/slide, and click Predict....

Step-by-Step User Workflows

1. Submit an Inference Run (from Viewer)

  1. Open a slide in Viewer.
  2. Right-click either a selected ROI annotation or inside the slide canvas and click Predict....
  3. Choose an available model (with matching preprocessing/class mapping checks) and confirm submission.

2. Monitor Inference Progress Across Projects

  1. Switch to Inference in the left sidebar.
  2. Browse the Recent Inference Runs table (inferenceJobsTable), sorted newest first by Started timestamp (sortDescending: true, 10 rows per page).
  3. Use Search inference runs... to filter runs by project name, slide name, model name, status, or detail string.
  4. Once Status shows completion, return to Viewer and enable the Predictions overlay in ViewerLayerToolbar to inspect polygons/heatmaps or export to QuPath GeoJSON.
flowchart LR
    ViewerPredict["Viewer -> Right-click -> Predict..."] --> Dispatch["Dispatch Inference Job (Local or Worker)"]
    Dispatch --> Monitor["Inference Tab: Recent Inference Runs Table"]
    Monitor -->|Running / Pending| Poll["Auto-updating Status & Details"]
    Monitor -->|Completed| Results["Prediction Set Saved on WSI"]
    Results --> Overlay["Viewer -> Predictions Layer & QuPath Export"]

UI Reference: Recent Inference Runs Table (inferenceJobsTable)

Column Key Header Label Width / Alignment Content Description
project_name Project Flex fill Name of the project containing the target WSI
wsi_name WSI 180px Slide filename analyzed
model_name Model 160px Registered model name/version used for prediction
status Status 110px (badge) Colored status badge (statusColor)
details Details 220px Execution telemetry, progress summary, or error reason
started Started 150px (monospace) Start timestamp; default sort key (descending)

Table configuration notes:

  • Pagination: pageSize: 10.
  • Search: Instant search via "Search inference runs...".
  • Read-Only Telemetry View: Row edit/view/delete buttons, manual filter dropdowns, and selection checkboxes are disabled on this monitor view (showViewAction: false, showEditAction: false, showDeleteAction: false). Cancellation or force-stop actions belong in Tasks, and prediction layer cleanup belongs in Viewer.

Technical Reference

  • Preprocessing Parity: Inference enforces identical normalization, tile size/MPP, and class label mapping as recorded in the model’s portable package manifest.
  • Separation of Prediction Runs: Multiple inference runs against the same WSI remain cleanly separated in prediction sets so users can toggle or delete a single run without affecting manual ground-truth annotations.