Skip to content

Models Tab

Overview

The Models tab (ModelsPage.qml, route Models) provides the Model Registry, unifying locally saved semantic-segmentation models and remote server-published model versions in one searchable catalog.

Model Registry table displaying Model Name, Version, Architecture, Encoder, Classes, Metrics, Created, Status, Location, and View/Download/Delete/Share actions.

Availability & Prerequisites

  • Browsing Registry: Available whenever modelsViewModel is initialized (shows "Model library unavailable" placeholder if not signed in / configured).
  • Importing Packages (.zip): Requires an active open project (hasActiveProject).

Step-by-Step User Workflows

1. Inspect Model Details & Provenance

  1. Click the View (eye) icon on any row in modelsTable.
  2. ModelDetailsDialog opens with full architecture metadata, encoder backbone, semantic class mappings, validation metrics, creation timestamp, and storage provenance.

2. Download Remote Models or Import Local .zip Packages

  • Download from Server: On any remote row where can_download is enabled, click the Download action (editIconName: "download") to pull the verified model archive onto your workstation for local inspection or inference.
  • Import Portable Package: Click Import package (.zip) (importModelButton) in the top-right header, pick a verified DeepHistoImageAI model .zip archive, and register it into the active project.

3. Share Models Across Organizations & Delete Local Models

  • Share selected: Select a model version (selectedModel.version_id) and click Share selected (shareModelButton) to open ModelSharingDialog, toggling organization/published visibility permissions.
  • Delete: Click the Trash icon on a local row (can_delete === true) to confirm deletion in deleteConfirmDialog. Remote/registry models cannot be deleted from this local action ("Only locally stored models can be deleted.").
flowchart LR
    Train["Completed Training / Local Archive"] --> Import["Import package (.zip) or Sync Registry"]
    Import --> Table["Model Registry Table"]
    Table --> View["Inspect ModelDetailsDialog"]
    Table --> Download["Download Remote Model -> Local Cache"]
    Table --> Share["ModelSharingDialog (Publish / Organization Scope)"]
    Download --> Infer["Available for Predict... in Viewer"]

UI Controls & Table Columns Reference

Header Actions

Button Object Name Disabled Condition Action
Share selected shareModelButton (Share selected) Disabled when !selectedModel.version_id Opens ModelSharingDialog for the selected version_id and published state
Import package (.zip) importModelButton (Import package (.zip)) Disabled when !hasActiveProject or modelsVm.busy Opens file picker to import a portable .zip model package

Model Registry Table (modelsTable)

Column Key Header Label Width Description
name Model Name Flex fill Human-readable model identifier
version Version 80px Semantic/registry version identifier
architecture Architecture 110px Segmentation family (e.g., U-Net, DeepLabV3+, SegFormer-B2)
encoder Encoder 110px Backbone encoder (e.g., resnet18)
classes Classes 180px Target semantic class names expected by the model
metrics Metrics 150px Summary validation scores (e.g., Dice / IoU)
created_at Created 130px Timestamp of registration/training completion
status Status 100px Colored status badge
location Location 180px Storage origin (Local, Server, or combined availability)
actions Actions Row action icons View (eye), Download (download), Delete (trash-2), plus table Refresh
Dialog Component Purpose
Model Details ModelDetailsDialog Full provenance, classes, training parameters, and metrics breakdown
Model Sharing ModelSharingDialog Manage organization sharing / publication state for a version ID
Delete Model deleteConfirmDialog Confirms irreversible deletion of a locally stored model version

Technical Reference

  • Archive Integrity & Runtime Bindings: Portable .zip model packages carry schema versions, manifest metadata, normalization parameters, class ordering, and pinned plugin runtime bindings so inference matches training preprocessing exactly.
  • Deletion Constraint: Deletion (can_delete) applies strictly to local disk copies; server-side publication and quota cleanup are governed by tenant permissions and cloud storage lifecycle rules.