Workflow

Export & deploy

The last stage of the loop: get your trained model and labeled data out of VisoLabel in the format you need, then run it anywhere.

Exporting a model

Open a finished training run and click Export. Choose a format for your target runtime:

Exporting your dataset

You can also export the labeled data itself — useful for version control, sharing, or training elsewhere. Supported dataset formats:

• YOLO — images plus txt labels and a data YAML.
• COCO — a single JSON with all annotations.
• Pascal VOC — per-image XML files.

VisoLabel can split the export into train/validation/test sets for you.

Running inference

You don't have to leave the app to use a model. Inside VisoLabel you can:

• Run the trained model on new images or video right in the project.
• Feed its predictions back into annotation as a starting point — a fast way to label the next batch of data.

Deploying elsewhere

Once exported, your model is a standard artifact you can deploy like any other:

• Edge / on-device — ONNX or TensorRT on cameras, Jetson, and embedded boards.
• Server / cloud — load the PyTorch or ONNX weights in your own service.
• Apps — bundle the ONNX model into desktop or mobile applications.

A minimal ONNX inference call looks like this:

import onnxruntime as ort
import numpy as np

session = ort.InferenceSession("visolabel-model.onnx")
outputs = session.run(None, {"images": image_batch})
# post-process outputs into boxes / masks

Improving the model over time

Deployment isn't the end — it's a loop. Collect the cases your model gets wrong in the field, bring them back into a project, label them, and retrain. Each cycle makes the model better on your real-world data.

You've completed the full workflow. Head back to the overview to revisit any stage.