Skip to content

Get started

Quickstart

This guide takes you from a folder of raw images to a trained model in about ten minutes. It assumes you've already installed VisoLabel.

1. Create a project

Open VisoLabel and click New project. Give it a name and choose a task type:

  • Object Detection — bounding boxes around objects.
  • Instance Segmentation — pixel-perfect polygons.
  • Classification — a single label for the whole image.

It sets sensible defaults for annotation and training. More detail in Projects & data.

2. Import your data

Drag a folder of images or videos onto the project window, or use Import → Add files. VisoLabel copies references to your files — nothing is uploaded anywhere. See Projects & data for supported formats.

3. Define your labels

Open the Labels panel and add a class for each thing you want to detect (for example cat, dog). Each label gets a color and an optional hotkey for fast switching while annotating.

4. Annotate

Now the fun part. You have three ways to label, fastest first:

  • SAM3 auto-annotation — let the model pre-label the whole dataset, then review.
  • SAM2 assisted clicks — click an object and SAM2 produces the polygon for you.
  • Manual — draw boxes or polygons by hand.

For a 10-minute run, start with SAM3 auto-annotation and fix only the mistakes. Full details in the Annotation guide.

5. (Optional) Augment your data

Got a small dataset? Click Create Augmented Dataset to multiply your examples — rotations, flips, color shifts, and more — without labeling anything new. A ×3 multiplier turns 2,000 images into 6,000. See Datasets & augmentation. You can skip this on your first run.

6. Train a model

Once you have a few dozen labeled images, click Train:

  1. Pick the dataset and adjust the train / validation / test split.
  2. Choose a model — start with RF-DETR-N for speed.
  3. Set the number of epochs — 50 is a good first run.
  4. Click Train locally (your GPU) or Train in Colab.
  5. Watch the live metrics as it trains.

More options in the Training guide.

7. Export & deploy

When training finishes, open Export to download your model weights in ONNX, PyTorch, or TensorRT format — ready to run anywhere. See Export & deploy.

That's the whole loop. To go deeper on any stage, follow the workflow overview.