01
Architecture and data handling
Ask where source footage, temporary files, telemetry and support diagnostics go. Distinguish local processing from a web interface that still uploads video.
Confirm licence validation behaviour, update delivery and whether the application can operate under the organisation’s endpoint controls.
02
Quality and review controls
Avoid relying on a single accuracy percentage without a dataset definition. Test representative profiles, crowds, low light, reflections and scene cuts.
- Can operators add missed faces?
- Can false tracks be removed?
- Are gaps and uncertain moments surfaced?
- Can the final export be reviewed before release?
- Does the tool preserve audio and required resolution?
03
Commercial fit
Compare licence seats, device activation, minute fees, storage and support. Model high-volume months rather than only an average job.
Define an acceptance test and exit criteria before the pilot starts.
Apply the workflow
Review real footage locally.
CensorFlow detects and tracks visible faces on your Windows PC, then lets a person correct and approve the result before export.