01 / live demo / 2,315 photos / 21 tickets each / 4 checks
Twenty-one tickets a photograph, three photographs a second.
Tickets come off this press 21 at a time, and each one carries a small printed mark called a Q-block. When the print starts to fail, those blocks are the first thing to go faint, go missing, or drift out of place. EagleEyes photographs all 21 at once and puts them through four checks. Fail one, and the whole photo is rejected.
Nissha Medical Technologies · EXT705 ticket press · replaying the real system’s own results
This replays real photos, slowed to a seventh of full speed and mixed the way a normal run looks: about 97 good in every 100. Each photo holds twenty-one tickets, and every pass or fail below is what the real system decided for that photo.
The press needs a screen with some room on it and motion you have not asked to reduce. Every measured result is below, and the stations are listed with it.
Five stations. Pick one.
EagleEyes, built for Nissha Medical Technologies on the EXT705 ticket press. Every result here is the real system’s own, from a run of 2,315 labelled photos. That test set was mostly bad photos on purpose, to prove the catches, so here they are shown in the mix a working press produces. Nothing else was changed.
02 / how well it actually did
2,315 photos, nothing missed and nothing wrongly rejected.
Every photo came labelled good or bad by the camera system already on the line, so each verdict can be checked against that label. Over all 2,315, it agreed every time.
192
good photos passed
of 192, 100.0%
2,123
bad photos caught
of 2,123, 100.0%
0
good photos wrongly rejected
waste the press would have paid for
0
bad photos let through
the expensive kind of mistake
Read that with care. The pass marks were measured from good photos and then adjusted until good print passed, and those same good photos are scored here. All 2,123 bad photos came from one hour of one night, so this shows it catches that batch’s fault. A fair test needs good and bad print from many runs.
03 / the four checks
Finding the blocks is half of it. Four checks decide.
Each check asks one simple question, with a pass mark measured from good print and then tuned by hand. All four have to pass. One catch: as delivered, the position check looks for its limits in the wrong place in the settings, so on its own it never fails a photo.
| Check | What it asks | Passed |
|---|---|---|
| Count | Are all the blocks there? | 2,297 / 2,315 |
| Visibility | Is each one crisp enough to be sure of? | 192 / 2,315 |
| Density | Is there enough ink in it? | 192 / 2,315 |
| Position | Is it in the right place relative to its neighbours? | 2,297 / 2,315 |
The exact pass marks, for anyone who wants them
- Count
- exactly 14 or 21 big blocks
- Visibility
- confidence 0.910–0.940
- Density
- mean grey 38.3–65.58, dark ratio 0.6484–0.9297
- Position
- within 3σ of the learned spacing
Where the time goes
Finding the blocks takes 79.2 milliseconds a photo. The four checks take 4.2. So the checks are almost free, and nearly all the effort goes into finding the blocks.
What went wrong, and how often
- Block too faint to trust2,105
- Not enough ink in the block2,105
- Wrong number of blocks17
- Nothing found on the ticket1
The first two nearly always happen together. A block that has lost ink is both fainter and harder to be sure of, so two checks catch the same problem from two sides.