Freight intelligence runs on cameras, but most camera networks only read plates and DOT numbers. We asked whether a cheap decision model can add equipment attributes on top: what the truck is hauling, and whether it shows a hazmat placard. Trio-Spark saw each photo once and returned a probability for every option.
77 public photos of US trucks from Wikimedia Commons, each checked by hand: one clear truck per photo, mixed angles, distances, weather and lighting.
| Labeled as | Correct | Mistaken for | Status |
|---|---|---|---|
| Dry van | 8/9 | container ×1 | solid |
| Reefer | 4/11 | dry van ×7 | weak |
| Flatbed / lowboy | 8/11 | dry van ×2, dump / bulk ×1 | mixed |
| Tanker | 15/15 | — | solid |
| Car hauler | 6/6 | — | solid |
| Dump / bulk | 12/13 | tanker ×1 | solid |
| Container | 11/12 | dry van ×1 | solid |
Chance for 7 options is 14%. Options offered to the model: dry van, reefer, flatbed or lowboy, tanker, car hauler, dump or bulk, intermodal container on a chassis.
Misses first. Each card shows the model's top answer and its probability; the credit line links to the original photo.
Same photos, second question: is a diamond-shaped hazmat placard visible? Below are every photo that has one, then every false alarm. Both false alarms are empty placard holders on the back door, a diamond outline with nothing in it.
Send about 10,000 labeled photos from your own cameras. We return accuracy per attribute (equipment type, loaded vs. empty, placard present, and logo-to-carrier match) on your data and your camera angles. If an attribute falls short, as reefer does here, that is the set we work on next.