Need Something
Detected?

Make the target object the main focus

Keep the objects you want detected large, clear, and prominent in the image or video. The larger and more visible the object is, the easier it will be to identify and label accurately.

Use a suitable resolution

Use 1920x1080 resolution when possible. Higher resolutions, including 4K, are acceptable but usually unnecessary. Lower resolutions may not provide enough detail for accurate labeling or fine-detail detection.

Include both good and defective samples

For defect-detection challenges, provide examples of both acceptable and defective products. The model needs to understand what "good" looks like as well as the defects it should identify.

Submit multiple images or videos

Whenever possible, include several images or video clips. This lets us reserve some samples exclusively for testing rather than using everything during training, giving a more reliable proof of performance.

Use a representative camera

Try to capture your samples with a camera similar to the one you plan to use in the final setup. For example, if the production system will use a fisheye lens, it's best to use that type of camera from the start.

Keep the camera steady

Phone videos work great, but please keep the camera steady whenever the target object is visible. We can remove unnecessary footage, but can't accurately label objects that remain blurry, obstructed, or out of focus.

Prefer short, focused video clips

Several short clips with useful activity are better than one long recording with extended periods where nothing happens. Focused clips are faster to review, label, and evaluate.

Provide clear and complete descriptions

Describe the challenge, target objects, expected behavior, and important conditions as clearly as possible. Your written notes are often the primary source of information for the person completing the challenge.