Annotate from scratch without an AI model

When you are starting a new inspection project — or setting up a new camera position — there is no trained model yet. Every image needs to be labeled by hand. This guide walks through the complete workflow, from finding the right images to saving annotations.

Note

Annotations you create here become the training data for the first model. Accurate, consistent labeling at this stage directly shapes how well the model performs later.

Overview

The workflow is:

1. Filter images on the Home page to find what you want to annotate
2. Build an annotation queue
3. Work through images one by one, labeling each one
4. Submit and move to the next image

Step by step

1. Open the Home view. Navigate to the Home page. Set the time range and any other filters to find the images you want to annotate.

2. Filter for unannotated images. In the filter pane, set Show / hide annotated to “Unannotated only”. This ensures you only see images that have not been labeled yet. Click Apply filters.

3. Select images to annotate. Either click Select displayed to select all images on the current page, or use Select all to select everything matching your filters. You can also select individual images by clicking the circle button on each card.

4. Start the annotation queue. Click Annotate selected images in the filter pane. EyVz opens the annotation view and loads your selected images as a queue.

5. Review the image. The annotation view shows the full-size image. Use the brightness and contrast controls if needed to see details clearly. The image canvas supports zoom and filters for difficult cases.

6. Set labels for each annotation. Below the image, you see the annotation form. Each label has a set of radio buttons — the available options. Since there is no model yet, no predictions are shown. Select the value that matches what you see in the image.

   For object detection labels, draw a bounding box on the image around the area of interest before selecting the label value.

7. Submit the annotation. Click Submit to save your labels. EyVz automatically advances to the next image in the queue.

8. Continue through the queue. Repeat for each image. You can use Shift+Right Arrow to move to the next image and Shift+Left Arrow to go back.

Tip

If you are uncertain about an image, you can click Skip Annotation to move past it without labeling. Skipped images are hidden from future annotation queues by default, but you can find them again by disabling Hide skipped annotations in the filter pane.

Tips for effective annotation

• Be consistent. If two images show the same condition, label them the same way. Consistency matters more than perfection for training data.
• When in doubt, skip. A wrong label teaches the model the wrong thing. It is better to skip an ambiguous image than to guess.
• Work in batches. Annotating 50–100 images at a time helps you stay focused and maintain consistency.
• Use tags to organize. After annotating, you can tag images (for example, reviewed or edge-case) to help with later review.

Deleting an annotation

If you made a mistake on an image you already submitted:

1. Find the image on the Home page. Filter by time or use “Annotated only” to narrow results.
2. Click the image thumbnail to open it in the annotation view.
3. Click “Delete Annotation” to remove the existing labels.
4. Re-annotate the image with the correct values and submit again.

What’s next?

Understand model behavior Once a model is trained, learn how to evaluate its predictions.

Home view Detailed reference for all filters and controls.

Glossary Definitions for terms used in EyVz.