# What are false positives and false negatives in egg grading? > A false positive is a clean egg that the system rejects as dirty or cracked. A false negative is a defective egg that passes as clean. The two errors cost different things, and reducing one usually raises the other. Vigiovo balances that trade-off with fixed thresholds and a 2-of-3 check vote. Updated: 2026-10-09 · URL: https://vigiovo.com/en/resources/false-positives-and-negatives-in-egg-grading · Português: https://vigiovo.com/recursos/falso-positivo-e-falso-negativo.md ## Key takeaways - A false positive is a clean egg rejected; a false negative is a defective egg that passes as clean. - Both errors can only be measured against eggs labelled by a trained person, not against someone's impression of the line. - Stricter thresholds usually raise false positives, and looser ones usually raise false negatives. - The 2-of-3 vote removes false positives from a single frame, such as glare. - In Vigiovo's published synthetic tests, no clean egg was rejected. ## What are false positives and false negatives? A false positive is a clean egg that the system judges dirty or cracked. A false negative is a defective egg that the system judges clean. Both are errors against the correct label, so the first question is who sets that label. In practice the label comes from a trained person looking at eggs under your light and on your line. A faint mark can be dirt to one operator and natural shading to another, so the system has to follow a single agreed rule. **The two errors and their costs (practical assessment)** | Error | What happens | Typical cost on the line | | --- | --- | --- | | False positive | A clean egg leaves as dirty or cracked | A good egg lost and product wasted | | False negative | A dirty or cracked egg leaves as clean | A defect reaches packing, with risk of quality and audit problems | ## Why does reducing one error usually raise the other? A system that decides by threshold has to choose where to cut. A stricter threshold catches more defects, but also rejects more clean eggs with similar-looking marks. A looser one protects good eggs, but lets more slight defects through. There is no point without a cost. The right choice depends on what your operation considers more expensive: losing good eggs or releasing a dirty one. That decision belongs to the quality team, and the system only applies it in a fixed way. For that reason, any accuracy figure should come with the conditions of the test. A number without your eggs and your light says little about your line. ## How does Vigiovo reduce false positives? Vigiovo uses three measures against false positives. Dirt is measured against each egg's own shading surface, so the natural shading of the shell does not count as a stain. Glare is excluded from the dirt measurement. And on video, an egg is rejected only if the defect shows up in 2 of 3 checks. The vote is the most direct defence against a false positive from a single frame: a reflection that appears in one check is not enough to reject a clean egg. The details are in [two-of-three voting](https://vigiovo.com/en/resources/two-of-three-voting) and [glare and false dirty grades](https://vigiovo.com/en/resources/glare-and-false-dirty-grades). ## How are false negatives reduced? False negatives depend more on what the camera can see. A crack only counts if it is elongated, long and deep, so a defect that does not meet those criteria is not counted as a crack. For doubtful cases, Vigiovo prefers not to judge. An egg cut by the image border, overlapping another egg or with an unusual shape is marked as an [unverified egg](https://vigiovo.com/en/resources/unverified-eggs) and gets no class by guesswork. That keeps the uncertainty visible, instead of turning it into a clean egg that nobody checked. Some defects no surface camera can see. Hairline cracks and interior defects need [candling](https://vigiovo.com/en/resources/glossary#candling) or acoustic testing, as the comparison in [candling vs computer vision](https://vigiovo.com/en/resources/candling-vs-computer-vision) explains. A surface system does not remove that kind of false negative. ## How do you know whether the system errs on your line? Vigiovo's published accuracy figures come from synthetic scenes with known ground truth, not from a production line. In those scenes, per-egg accuracy on tracked video was 376 of 381 eggs, and no clean egg was rejected. Single-frame per-egg accuracy was about 95%. These figures show how the method behaves, but they do not replace measurement on your line. To learn the false positive and false negative rates of your operation, set aside a set of labelled eggs, measure both errors and adjust the thresholds one at a time. The step-by-step is in [calibrating with labelled eggs](https://vigiovo.com/en/resources/calibrating-with-labelled-eggs). ## Frequently asked questions ### Is a system with no false positives possible? Not in practice. Any threshold that separates clean eggs from dirty ones has borderline cases, and accepting fewer errors on one side usually raises the error on the other. What you can do is choose the right balance for your line and measure both errors with labelled eggs. ### Is a false negative worse than a false positive? It depends on your operation's costs. A false negative lets a dirty or cracked egg through, which can cause quality and audit problems. A false positive throws away a good egg. For that reason the quality team should set the threshold, not the system alone. ### Do Vigiovo's published figures apply to my line? Not directly. They come from synthetic scenes with known ground truth. Your camera, lighting and egg colour change the result, so accuracy on your line is only known after calibrating with real images. ## Sources - [Como o Vigiovo classifica ovos sem IA / How Vigiovo grades eggs without AI](https://vigiovo.com/como-funciona), Vigiovo ## Read next - [Calibrating with labelled eggs, step by step](https://vigiovo.com/en/resources/calibrating-with-labelled-eggs.md) (Guide, 2026-10-09): A step-by-step guide to calibrating Vigiovo with a few hundred labelled eggs: what to label, which limits to tune and how close each decision sits to its threshold. - [Two-of-three voting: why measure each egg three times](https://vigiovo.com/en/resources/two-of-three-voting.md) (Explainer, 2026-10-09): How Vigiovo measures each egg in three checks as it passes on the conveyor and rejects it only when the defect shows up in two, which filters out isolated glare. - [Candling vs computer vision: what each one detects](https://vigiovo.com/en/resources/candling-vs-computer-vision.md) (Comparison, 2026-10-09): Candling compared with camera inspection in egg grading: which defects each detects, speed, record keeping and why the two methods work best together. ## How Vigiovo helps Vigiovo grades eggs on the conveyor as clean, dirty, cracked or unverified, in real time, with deterministic computer vision: every decision comes with the measurements behind it, on an ordinary CPU. https://vigiovo.com/en/how-it-works.md 4 min read.