Guide · Running the line
How do you set up a computer vision pilot for grading eggs?
A camera inspection pilot measures, on your own line, how accurate the system is. Set up a fixed camera above the conveyor with diffuse light, label a few hundred eggs with your team, run the classifier on an ordinary computer and compare each decision with its label. The outcome is an accuracy report and a configuration calibrated for your line.
Updated · 4 min read
What does a computer vision pilot need?
A pilot needs four things: a fixed camera above the conveyor, diffuse lighting, a few hundred eggs photographed and labelled by your team, and an ordinary computer near the line, with no GPU. With those in place, the result is an accuracy report measured on your line and a configuration calibrated for it.
Before you start, check each item against the pilot readiness checklist. It also helps you name who reviews the eggs without a clear decision, which needs to be settled before the start.
How do you set up the camera and the light?
The setup comes before the software. Fix the camera above the conveyor and keep its position throughout the pilot, because any shift changes the measurements. The light must be diffuse, with no strong glare on the shells, since glare is the main source of false dirty eggs.
A basic industrial camera, or even a phone, is enough to start, and high resolution is not needed. For more on the light, see the guide on lighting for egg inspection.
How do you label the pilot eggs?
The labels are the ruler of the pilot. A trained person classifies each egg as clean, dirty, cracked or unverified, without looking at the system's decision. Deliberately include the doubtful cases, such as light stains, fine cracks and eggs touching the edge of the image, because they show where the limits should sit.
Set aside a test set before you calibrate, and do not use it to tune limits. The tuning steps are in calibrating with labelled eggs.
How do you compare the system's decisions with the labels?
Run the calibration set and compare, egg by egg, Vigiovo's decision with the label. Each decision shows the measurement and the limit side by side, so the most instructive errors are the eggs that sat close to the limit. Errors usually have different causes, and each one calls for a different first step:
| Error | What shows up | First step |
|---|---|---|
| False dirty | A clean egg marked dirty | Check glare and shadows before changing the dirt limit |
| False clean | A dirty or cracked egg marked clean | See whether the stain or crack fell below the limit |
| Unverified | An egg cut by the edge, overlapping or of abnormal shape | Adjust the framing so the eggs appear whole |
Change one parameter at a time and run the same set again to see the effect of each change. Because classification is deterministic, the same image with the same configuration always gives the same result, so the comparison can be repeated whenever you need.
What should the pilot report show?
The report shows the accuracy on your line, by class, measured against the labels of the test set. It also shows how many clean eggs were rejected and how many were left as unverified for manual review.
The report should also say what the camera cannot see. A camera pilot does not see the air cell, the yolk, internal blood spots or hairline cracks, which need candling or acoustic testing.
The accuracy figures Vigiovo publishes come from synthetic scenes with ground truth, not from a production line. The number that matters for your farm or grading station is the one in your own report, measured on your images. The full approach is explained in how it works.
Frequently asked questions
How long does a pilot last?
- That depends on how long your team takes to label the eggs and how many calibration rounds the line needs. Define the end of the pilot by the success criteria agreed before the start, not by the calendar.
Who should label the pilot eggs?
- The labelling should be done by a trained person from your own team, who knows the eggs on your line. The result of the pilot is an accuracy report and a configuration calibrated for your line, measured with those labels.
Does the pilot show whether a camera can replace candling?
- No. A camera cannot see hairline cracks or the inside of the egg. The pilot measures what the camera does well on your line, which is dirt and visible cracks, and leaves eggs without a clear decision for manual review.
Sources
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.