RFD 1143: The keypoints-to-ANNY loop
Problem
fourloops-plan.usda carries four loops as prims, each with its stages, its artifact and one line of hazard. A prim holds the wiring and cannot hold the argument, so the reason a loop is shaped this way lived in a logbook entry that the plan named as a source.
All four loops are wanted. What was missing is a document per loop: the plan holds the wiring for four, and one logbook entry held the argument for all of them. This RFD is loop 1’s own.
Decision
Propose with Keypoints, AnnyFit and Render; score with EditScore and the Referee; repair by refitting. The artifact is a render png. It is the largest of the four at size L, and it runs fourth in the task order because the harness comes first.
Two scorers rather than one, because they answer different questions. EditScore reads the image and the Referee reads the body, and a fit that looks right while the joints are wrong passes the first alone.
The hazard is the joint count. The detector emits 17 keypoints and the render asset emits 23. A comparison across those two takes an explicit subset, or it compares different things and reports a number for it. RFD 1122 records the same gap for the wholebody chain.