Skip to main content
Key frames are a faster way to track an object across a sequence. Instead of positioning a cuboid on every frame, you position it on a handful of key frames and Avala fills in everything between them. If the object moves smoothly, a 200-frame sequence might need five or six key frames.
Key frames are being rolled out gradually. If you do not see the options described here, your account does not have them enabled yet — nothing you have already annotated is affected.

​
The two kinds of frame

You can tell them apart on the timeline strip under the viewer:

​
Annotating a sequence

  1. Choose Add cuboid (key frames) when you start a new object.
  2. Go to the first frame where the object appears and place the cuboid. Size it and set its heading as usual.
  3. Jump ahead — to where the object is next clearly visible, or to the last frame it appears on — and place the cuboid on the object there.
  4. Press the ← button on that second key frame to join it to the one behind it. The frames in between fill in immediately.
  5. Scrub back through the middle and check the fill. Wherever it has drifted off the object, just fix it. Adjusting a derived frame turns it into a key frame automatically, and the frames around it re-fit to match.
Repeat step 5 until the cuboid tracks the object everywhere. You are not aiming for a fixed number of key frames — add one wherever the automatic fill is not good enough, and nowhere else.
Step 4 is not optional. Two key frames sitting next to each other do not fill the gap on their own — Avala never assumes that two poses belong to the same continuous stretch of movement, because sometimes they do not. You say they do, with ← or →. See When the object disappears for why that is the safer default.
Add key frames where the motion changes — the object starts turning, brakes, pulls away. A long straight stretch needs a key frame at each end and nothing in between.

​
Marking and unmarking by hand

Press M, or use the diamond button in the toolbar, to switch the current frame between key and derived.
  • Marking a derived frame pins the pose exactly where it is. Nothing moves.
  • Unmarking a key frame hands it back to Avala, so it re-fits from its neighbours. The cuboid will usually shift a little — that is the point.
The button’s tooltip tells you how many frames the change affects before you commit to it, and the bar above the viewer names the range your current edit will re-fit — for example “Key frame — edits re-derive frames 12–48 (35 frames)”. The first and last key frames of a track cannot be unmarked; there is nothing beyond them to fill from. To shorten a track, delete the cuboid from those frames instead.

​
When the object disappears

Objects go behind things, leave the frame, and come back. Avala should not invent a path across a gap it cannot see. The ← and → buttons on a key frame say whether it and its neighbour are one continuous stretch of movement:
  • Connected (highlighted) — the frames between this key frame and its neighbour get filled in.
  • Disconnected — they do not. Any frames already in that gap are removed, and the object is treated as two separate stretches.
New key frames start disconnected, which is why step 4 above exists. Avala will not invent a path across a gap you have not vouched for — if an object vanishes behind a building and reappears further along, the straight line between those two poses would run through the building. Connect when the object was there the whole time, even if you could not see it clearly: a cuboid that follows an occluded car is usually what you want. Leave it broken when the object genuinely was not there. The tooltip tells you how many frames you are about to add or remove before you press it.

​
Moving an existing track onto key frames

Tracks made the older way keep working exactly as before; nothing is converted behind your back. If you have the option, To key frames in the toolbar switches the selected track over. Every frame stays exactly where it is — nothing moves — and every frame starts out as a key frame. From there, unmark the frames you would rather have filled in automatically. There is no button to switch back, so it will ask you to confirm. Undo still works.

​
Common mistakes

  • Forgetting to connect. If you place two key frames and nothing fills in between, you have not pressed ← or → yet. The timeline gives it away: two lone diamonds with no bar between them.
  • Placing a key frame on every frame. That is the old workflow with extra steps. Place a few, check the fill, and only add more where it is wrong.
  • Fighting a derived frame instead of fixing it. If a filled pose is wrong, drag it. It becomes a key frame and everything settles around it.
  • Breaking a run for an occlusion. If the object is still there but hidden, keep the run connected. Break it only when the object has actually gone.
  • Not checking the middle. After adding a key frame far ahead, scrub back through the frames between. That is where the fill either worked or did not.

​
Shortcuts