Skip to content

The Cleanup Plan ​

A take is one recorded motion-capture performance. Every take gets its own Cleanup Plan: the steps Post will take on it, in the order it will attempt them, and what each step is waiting on. The plan is built from that take's diagnosis, which is the read-only pass Post makes over the file when you upload it, and from the options you set. Two takes therefore never get the same plan.

It is the same list at three moments. Before the run it is a forecast. During the run it lights up. After the run it is a record. It fills in as work happens. It is never replaced by a second, different list.

Before the run ​

Open Clean, choose Assisted, and tick a take. Clean is the screen where you launch a cleanup, and Assisted is the mode where you choose the options for each take yourself. A third column then appears, headed Cleanup Plan. It says:

What we will do to this take, in order. Derived from its diagnosis, so every take is different.

Each row in that column is one kind of work, such as "Fill small gaps" or "Cut impulsive jitter". A row carries four things.

  • How much of it there is, in the noun that work is counted in: 62 gaps on 18 markers, 8 joins, 4 checks. Where the number of steps differs from the number of gaps, the row names both: 6 gaps in 4 steps on LT10.
  • What it sets out to close: 2 regions · 34 frames. A region is one span of frames on one marker that Post works on.
  • Why it runs here, written by the cleaner in plain words.
  • after: Fill small gaps, which names the steps that have to be attempted first. Rows are indented and connected, so you can see what waits on what without reading.

Some rows say counted as it runs instead of giving a forecast. Those steps count their own work as they do it and cannot know the number in advance, so the plan says so rather than guessing.

Changing a file or an option re-asks for the plan. While that happens, Updating the plan… appears in the column header. A plan can also be unavailable, and the column names which case it is. You may see Diagnose first to see the plan, This take will be re-diagnosed first, Nothing to plan, The plan is taking longer than usual, or The planner could not be reached. You can still launch the run, because the launch does not wait for the preview.

During the run ​

The column keeps the plan the run was launched with, and each step is marked with what happened when the clean finishes:

Waiting its turn. Each step is marked with what happened when the clean finishes.

Cleaning. Each step is marked with what happened when the clean finishes.

After the run ​

Two places now show the finished plan. The first is that same third column on Clean. The second is the Cleanup Plan block on the Review screen, which is the screen you read once a run has finished. Both now read:

What the run did to this take, step by step.

Every row gains what it actually did (1 of 2 regions · 9 of 13 frames) and a state badge. A folded row rolls its steps up in one line: 58 done · 4 not needed.

Above the rows, one line counts what the run left behind. There is only ever one such line, so you never have to reconcile two tallies. It reads as one of these:

  • 4 frames ship open · 1 step waiting for another run
  • Finished inside its 3 iterations, with nothing left waiting
  • Nothing needed, because no step had work to do
  • Another run would improve this take
  • the plan ran, but the open frames were not recorded, which means nothing counted the open frames, so no number is stated.

Underneath it, a second line tells you what to do about it. It reads Run again to try it, and Smooth joins goes first, or Another run would not change this, so hand these frames to an artist when a second run would schedule the same work and decline it the same way.

Each performer's row carries a one-word verdict: Complete, Nothing needed, Nothing open, Still waiting, Did not land, or No result yet.

Step states ​

There are four words for how a planned step ended. None of them is an error.

BadgeWhat it means
DoneThe step ran and put something in the delivered file. Less than it forecast is a normal result, and a Done step can still leave frames open, with the count beside it.
Not neededThere was nothing to do. An earlier step had already closed the region, or the option that would have run it was off, or the step looked at its windows and handed every one of them back with nothing left open.
Still waitingIts prerequisites were still not met when the last pass ended, so it ships undone. This is the state that says another run would buy you something, and a re-run attempts it first.
Did not landPost attempted it and shipped nothing. What it stood for is in the delivered file as holes, for an artist to finish.

Two more words show up in a folded row's rollup.

  • left open counts the units a step stood for and did not close. They are counted separately so that Done never claims work that is not in the file.
  • absorbed into a neighbouring fill follows the done count on a gap row, for example 3 done, 1 of them absorbed into a neighbouring fill. Two planned gaps came out as one fill, and the take was measured with no frames left open, so that gap is done rather than left open.
  • Reported is a state this build of the app does not recognise. The counts are kept and nothing is claimed about them.

"Did it land?" and "is anything still open?" are two different questions. The badge answers the first. The count beside it answers the second.

The steps ​

There are eight kinds of work. A plan uses only the ones a take needs.

StepWhat it doesCounted in
Fill small gapsCloses a gap from its own two anchors. Needs nothing else to be true first, so it usually goes early.gaps
Fill medium gapsCloses a longer gap using the trajectory around it.gaps
Fill long gapsSolves against the neighbouring markers, so it is planned after the neighbours it reads are clean.gaps
Cut impulsive jitterCuts an impulsive acceleration peak out (the peak plus a small pad) and refills the hole, the way you would replace a spike rather than average it into its neighbours. Only planned when the cut option is on.cuts
Fix swapped markersResolves a swap or a merge across both markers of the pair.swaps
Smooth jitterSmooths jitter that was not cut.jitter windows
Smooth joinsConditions the boundaries between what the run wrote and what it did not.joins
Check our own fillsRe-measures the result and decides whether another pass is worth running.checks

The order is not a fixed recipe. It is worked out per take, from what the diagnosis found and what each step needs to be true first. The steps are then grouped into passes, and everything in pass 2 may assume pass 1 was attempted. The after: line on a row names what that row waits on.

Why the jitter cut comes after the fills ​

A cut takes frames out and has to rejoin the trajectory on both sides. Next to a hole there is nothing to rejoin to, so the cut is planned after the fills of its own pass, once the surrounding data exists.

That is also why there are two cut passes. The first reads the take as captured. The second reads the bytes the fills produced, because filling a gap can expose a peak that was not visible before. A region card on Review says which pass a cut came from, in the words Pass 1, before the gap fill or Pass 2, over the filled result.

What the plan does not do ​

  • It is not a quote. The forecast on a step is what that step sets out to close, not a promise.
  • It does not say what the delivered file still has open. That is a separate, measured number. You will find it on the line above the rows, and in Holes left open.
  • It is not shown in Auto mode. Auto cleans every take in To clean with the standard fix order and no per-take column.
  • A Clean version made before plans existed has none, and the column says so rather than pretending the run did nothing.

What to do next ​

Read a plan that came back with Still waiting steps and run the take again, because those steps go first. Read one that came back Did not land with frames open and hand those frames to an artist, because another run will decline them the same way.

Ask us ​

Something looks wrong on a take? Write to support@kapnetix.ai and send the take id, or the take's name and the project it sits in.