Video Annotation You Can Actually See

TLDR. You define the taxonomy. We annotate video to it and send burned-in previews so you can judge quality with your own eyes.

You should not have to take annotation quality on faith. At posterior.xyz we annotate to your definitions and return video you can watch. If a box drifts during contact or a segment starts late, you see it right away.

Your taxonomy runs the labels

Every team means something slightly different by interaction or handling. We do not push a generic set. You bring your definitions and approved examples, and we write guidelines from them.

• boxes for hands and for tools in scope, plus people where your task calls for them
• segments that mark start and end of steps, with IDs that hold within a shot

We add attribute labels for state change and contact. When an edge case comes up, we flag it for your ruling and version the guideline so future work stays consistent.

Tracks that hold across time

Single frames miss motion. We follow subjects across time so identity holds through occlusion and handoffs. It also holds across viewpoint changes where your protocol allows. Kinda simple to say, harder to do well.

You tell us how to handle re-entry within a scene and camera movement indoors or outside. If your rig pairs video with synced depth, alignment stays intact across streams in HD to 4K.

Burned-in previews you can watch

Annotation files are exact and painful to judge by eye. So we render boxes and segments straight onto the footage. You press play and watch where a box sits during contact and when a segment starts relative to motion.

• reviewers who never open an annotation file can still judge quality
• engineers can cross-check the picture against the export

Review without our tooling

Previews play in any standard player. Exports use the format you specify. You sample, leave timestamped notes, and we correct against the same references.

1. You set acceptance criteria and sampling during the pilot
2. We send annotated batches with previews plus exports in your format
3. You review on your own and return timestamped notes
4. We fix and resubmit, with a guideline update if your ruling changed anything

Because guidelines carry version numbers, every change in interpretation traces back to a call you made.

Questions buyers ask

What do you need from us to start?

Your taxonomy definitions with approved examples. Your export format with a sampling plan for the pilot.

How do corrections work?

Send time stamped notes on the preview. We correct the labels and update the guideline version when your note changes the rule.

Start with a paid pilot. Both sides check fit before scale. Request a pilot or Talk to us.