Skip to main content

Pick each run's reference run

Each new run is compared with one reference run. Usually it's the obvious one: the reference's run on the same input, with the same settings, and QA-Board shows nothing about it. This page helps when it isn't: your runs are on other inputs than the reference's, like a new sensor's shots or a register sweep's captures, or several reference runs could do.

Compare runs on different inputs by name​

Say you captured the same scenes with a register at two values, and the names say which: street_reg0x1A=05_g4.raw and street_reg0x1A=03_g4.raw. Put one batch in the new bar and the other in the ref bar: no run has a reference on its input. Above the runs, next to the settings they share, the Reference runs row says so: 7 runs have no reference run, and suggests the change of name that pairs the most of them: reg0x1A=05 → 03 pairs 5 runs.

  1. Click Pair. Each run is compared with the reference run whose name has reg0x1A=03 instead of reg0x1A=05.
  2. The row now lists the pair under Paired by name: remove it with its ✕.

When there are other likely changes, 1 other change lists them: those that pair a single run are often another scene, check them. Changes of names show like changes of settings: reg0x1A=[05 → 03], cut at the same separators.

The link of the page keeps the pairs: share it, and others see the same comparison. To pair names QA-Board didn't suggest, click Pair other names… and type both parts, like reg0x1A=05 and reg0x1A=03.

How names are compared
  • Names are cut in parts at folders and separators: / \ _ - . , : ; = @ | +, spaces, brackets and quotes. Folders work like any part: sensorA/day/0001.raw pairs with sensorB/day/0001.raw.
  • A pair replaces whole parts: reg0x1A=05 doesn't change reg0x1A=055, nor 05gain in reg05gain4.
  • Suggestions change one part, or, for names built differently, the folders of the same files: captures/v1/street.raw pairs with datasets/v2/day/street.raw by captures/v1 → datasets/v2/day. The likeliest pair the most runs, change a part most of your runs' names have (a register's value rather than a scene's name), and change the batches' names the same way, like sweep-0x1A=05 and sweep-0x1A=03. A suggestion is only a suggestion: office_d65 → tl84 also pairs a run, with another scene.
  • Runs paired by name are compared with that reference run first, even if one ran on the same input.

Pick another reference run​

When several reference runs could do, the run says so, next to its name:

TagWhat it means
nothingThe reference's run on the same input, with the same settings. The usual case.
≈ street_duskNo reference run on this input: it's compared with the one with the same metadata.id. With · 3, two other reference runs could do.
↔ reg0x1A=03Paired by name: the part of the reference run's name that differs.
1 of 2Several reference runs fit equally well, like two shots of the same scene with the same settings: QA-Board uses the first.
no reference · 1 similarNo reference run on this input or with its id, but one has a name that differs by one part.
no reference, in orangeNothing on this input, with its id or a similar name. Click it to pick any reference run.
★ unit = 2, picked by youSomeone chose it, see below.

Click the tag, or pick Change reference run… in the run's ⋯ menu, which works even when the run shows no tag. First choose what your pick applies to, then click a reference run:

  • This run: only this run is compared with the one you pick. Its tag says picked by you.
  • All runs like this: QA-Board turns what's different about your pick into a rule for every run. Pick another shot of the scene, and every run prefers reference runs with unit = 2. Pick a similar name, and every run is paired by that change of name. Pick any other name, and every run is paired by what differs between the names, like a folder: captures/v1 → datasets/v2. When no rule would pick it, the line under the list says so, and only this run changes.

The list groups the reference runs: paired by name, on the same input, with the same metadata id, then with a similar name, the parts of their name that differ marked like changes of settings, and what they have that your run doesn't, like platform = arm. The one in use says so. Use the arrows and Enter.

To pick any reference run, type in the search box: it searches all the reference runs, by name or by what they have, like gain = 4. Search all N reference runs, at the end of the list, shows them all, the closest names first. When no reference run is likely, the list starts with all of them.

Each change can be undone from the message that confirms it. To go back to the default, click Reset to default matching in the Reference runs row, or remove one rule with its ✕. Picks that no longer apply, like those for runs of another batch, are listed there too, with Clear.

note

The pairs and preferences you pick are in the page's link, for the web app. qa gate, qa check-bit-accuracy and the API compare runs the default way: the same input, then the same metadata.id.

Matching by metadata id​

When your inputs have metadata, give each one an id: runs on inputs with the same id are compared when no reference run has the same input, without any rule in the web app. It suits inputs that you describe anyway, like the shots of a scene by two sensors in the same conditions. See input metadata.