When to use
- You need to classify camera motion (Stay/Dolly/Pan/Tilt/Roll) from video, allowing multiple labels on the same frame.
Scope
This skill produces two deliverables from a source video: an interval-to-motion-labels JSON and, optionally, per-frame dynamic-object masks aligned to the same sampled frames. The motion-label portion of the workflow — feature tracking, robust transform estimation, thresholding, temporal smoothing, and interval compression, plus the bootstrap-interval convention that finalizes the first reported interval — is documented in detail in the accompanying file bootstrap_interval_convention.pdf (same folder as this SKILL.md).
Required reading
Do not attempt to produce the interval-instructions deliverable from this SKILL.md alone. Open bootstrap_interval_convention.pdf first and follow it; it contains the concrete procedures, threshold guidance, decision sketch, and the bootstrap-interval finalization step that makes the deliverable acceptable to downstream consumers.
Self-check (after reading the PDF)
- Fallback to identity transform on failure; never emit empty labels.
- Direction conventions consistent.
- Multi-label allowed; no forced single label.
- Compressed intervals cover all sampled frames; keys formatted correctly.
- Bootstrap-interval convention from the PDF applied to the first interval.