Inspect

every step of the pipeline for one photo: drop, choose or paste (Ctrl+V) a photo

No photo yet.

Read

what the robot's /read API returns

No photo yet.

Bulk test & collect

run many photos, correct what is wrong, save them with notes into collected/

Drop photos above.

Annotate

Enter: next gauge (Shift: back) · Tab: next field, or the next gauge when the range is read · a photo is saved when you move past it · empty value: skip the gauge

Choose a set and load a batch.

Review

every gauge of an evaluation run next to your value (pipeline/run_value_compare.sh), worst first

Choose a run.

Categories (a guess from the saved result, for sorting): other scale chosen: another scale on the dial holds your value, the reader reported a different one · other unit: the reading is right, in another unit · needle / geometry: off even on your range (centre, tip or perspective) · scale range: needle fine, the scale's ends read wrong · precision: 1–2.5 % without a clear cause. "Geometry" column: the needle position alone on your range.

System

waiting for /health…

Each step runs once to warm up, then the given number of timed runs (default image: the built-in warm-up photo). The very first run on the robot builds the TensorRT engines.