Models

The Models section collects the AI models available to your account. From here you can create or import a model, open its dataset, start training, and download exported results.

Each model is displayed as a card with a dataset preview, statistics, and contextual actions.

Model Types

BIION-D Cloud supports the following model types. Each type is identified by a badge and follows a different workflow.

Model List

The main Models page shows a grid of cards. The upper toolbar contains:

• Upload model, to import a .zip file;
• the + button, with the Create model tooltip;
• Search models…, to filter the list by name.

The list displays six models per page. When a search has no results, the page shows No matching models.

Card Contents

Each card can display:

• a large dataset preview image;
• the model type badge;
• the model name, ID, and image count;
• a dataset summary bar for classified or annotated images;
• an orange dot when training has just completed;
• a status overlay while a job is queued, training, validating, or uploading results.

Card Actions

When you hover over a card, an action bar appears in the top-right corner.

The card training action appears after a requirements check. Unlike Train in the gallery, it starts the job directly without showing the detailed pre-flight summary.

If the required license is missing or expired, the affected controls are disabled and show License not available for this model type.

Training Status

Training status is displayed as an overlay on the card preview.

If another training run is active, the new job enters the queue and its position is updated automatically. The current interface does not provide a control for manually cancelling a queued job.

Model Detail

Selecting a card opens the gallery, which shows:

• the Models breadcrumb to return to the list;
• the model name and type;
• a License missing or expired warning when applicable;
• an image gallery initially displaying 12 images per page;
• Filter, View/Edit, Show Annotations, Annotate, and Train, when applicable;
• the history icon button, identified by the View training history tooltip.

For more details, see Image gallery and Annotation interface.

Training

Starting a Training Run

You can start training from the card with Train model or from the gallery with Train.

The gallery displays a Pre-flight summary with the required counts. Its status is Ready when all requirements are met or Incomplete when images or annotations are missing. In that case, Start Training remains disabled.

Requirements depend on model type and phase:

• AnomalyDetection: at least 20 good and 5 defected images;
• Object Detection and Object Segmentation: at least 5 annotated images;
• DefectClassification, first phase: at least 5 good and 5 defected images;
• DefectClassification, second phase: all defected images must be annotated.

Training Queue

If another training is already running, the job enters the queue. The card shows Queued… or Position N/M.

Training Error

If an error occurs during training, the card shows Training Failed and the training run is not completed.

Training History and Download

Creating a Model

Open the Creation Modal

Select the + button in the upper toolbar to open Create New Model.

Enter Model Name

Complete Model Name and select Model Type.

The selected type determines the workflow and compatible source models.

Select Source Models (optional)

Choose one or more Source Models to duplicate or merge their datasets. Each source displays compatible options such as Labels, Annotations, or Only images imported.

Sliders control the percentage of images to import, separating defected/annotated images from good/unannotated images when available.

Source → Target	What is Imported
AnomalyDetection → AnomalyDetection	Images + Compliance labels (optional)
YOLO → YOLO (same family)	Images + Annotations (optional)
AnomalyDetection → YOLO	Images only
YOLO → AnomalyDetection	Images only
AnomalyDetection → Defect Classification	Images + compliance labels — source model required

Creating a DefectClassification model requires at least one AnomalyDetection source. Labels are included, and the initial selection imports all defected images and a portion of good images. The first training phase starts automatically after creation.

Create

Select Create Model. When creation succeeds, Model created successfully! appears and the list refreshes.

Importing a Model ZIP

Select Upload model and choose a .zip file. Upload starts immediately and shows Uploading model…. When it finishes, Model uploaded successfully! appears and the page refreshes.

Next Steps

• AnomalyDetection — full workflow for anomaly detection models
• Object Detection and Segmentation — annotation and training workflow for YOLO models
• Defect Classification — complete workflow for the hybrid model
• Image gallery — browse, filter, and manage a dataset
• Annotation interface — create and edit bounding boxes, polygons, and masks
• Licenses and usage — review available features and usage
• Devices — connect and manage your hardware devices
• Workspace — configure workspaces for training