Skip to content

Train and export

Current release 1.9.0. This is the order of work on the engineering PC, from an empty project to an exported model.

  1. Install and license. Install Model Bench and confirm Help ▸ Licence… shows an active licence.
  2. Create a project and its classes. Each project has its own classes, images, review state and model versions. A project can hold more than one folder, so image sets stay separate. Classes can be renamed or merged later.

    Project dashboard: project list on the left, image gallery with status under each thumbnail, and the action buttons

    Classes dialog with three classes, each in its own colour 3. Add images. Upload Images or Upload Folder. Import COCO, YOLO or Pascal VOC if the set is already labelled. 3D Images Point Cloud Upload is for ToF captures (Pro). Use the folder filter to work on one image set at a time.

    Project dashboard filtered to one folder; the counts above the gallery change with the filter 4. Label and mark reviewed. Bounding box (B), polygon (P) or rotated box (O). Only reviewed annotations train the model. Press ? for the shortcut list.

    Annotation workspace: tools on the left, a selected rotated box with its rotate handle, and the annotation list on the right

    Keyboard shortcuts list, opened with ? or F1 5. Train. You need at least 50 reviewed images. The app says what is missing. Training uses an NVIDIA GPU when the PC has one, otherwise the CPU. The first line of the log says which. Stop Training finishes the current epoch and saves a checkpoint. Options chosen here (package corners, Hailo-safe) shape the model and cannot be added afterwards.

    Train Model page during a run: dataset check, split, model settings, and the live log naming the device and the warm start from the previous version

    Training log after a completed run, listing the images and classes the model was trained on 6. Auto-annotate, then review. Suggestions stay unreviewed until a person checks them. Correct them, mark them reviewed, and train again. A new run warm-starts from the previous weights when the classes match.

    Auto-Annotation page in loop mode: annotation progress, Train the Model, and Auto-Annotate Remaining Images

    An auto-annotated image open for review. Each suggestion is marked as needing review; Mark Reviewed accepts them 7. Export. ONNX is Standard. Hailo and NVIDIA GPU are Pro. The file carries class names and pre-processing. The export checks the converted model against the original and fails rather than ship a model that differs. Export on the dashboard writes the labelled dataset itself (COCO, YOLO, Pascal VOC or a full ZIP), for example to share it or to train elsewhere.

    Export ONNX dialog with the Target list open: ONNX Runtime, NVIDIA GPU, Hailo

    Export dataset dialog: format and an optional train, validation and test split 8. Bring production frames back. Visual Detector can save captures.zip. Import ctrlX Captures adds each frame as unreviewed annotations under ctrlx/<device>/cam<N>. Importing the same zip twice adds nothing. Correct, review, retrain, export.

Reference: editions, export targets, release notes.