Name the detail that matters
Specify what must remain visible: facial movement, a small hand-held object, scene text, or a full interaction. A large frame can contain a very small target. Request sample intervals showing typical and difficult framing, and inspect them at the size your pipeline will use. This grounds the requirement in usable evidence instead of an attractive resolution label.
Consider changing views
Video framing can change through camera motion, speaker switching, zooms, and edits. State whether the required detail must remain visible continuously or only during labeled events. Record intervals that fail the framing requirement rather than assigning one optimistic value to an entire recording. A single sharp frame should not establish that a long sequence supports a fine-detail temporal task.
Document delivery transformations
Ask whether the proposed export was cropped, resized, letterboxed, rotated, or enlarged from a smaller source. Preserve known original dimensions and version relationships. Check that bounding regions, face tracks, and temporal labels still refer to the delivered picture. If different aspect ratios are mixed, specify acceptable treatment rather than allowing a downstream importer to stretch scenes without an agreed policy.
Pilot with the intended workflow
Decode representative files, sample the relevant intervals, and inspect the result after preprocessing. Separate unreadable detail from file corruption so the response to each is clear. Discuss which constraints are mandatory and where a narrower crop or alternate source version might help. Fiund reviews the feasibility and commercial scope of those options for the particular proposed collection.
Review checklist
- Identify minimum task-relevant visual detail.
- Specify whether visibility must persist over time.
- Document crop, resize, and aspect-ratio handling.
- Review samples after intended preprocessing.