Look for human1. Create an operation and load footage
Name a local operation so findings do not mix with another task. Add photos or video with its matching SRT file. Keep original names and files: messaging apps and editors may strip telemetry. Prepare the model offline before deployment and test it on known footage.
2. Run AI review
D-FINE reviews visible imagery locally using available WebGPU or WASM. Thermal support provides separate auxiliary heat hints for operator assessment, not automatic person confirmation. Tiling increases review time. No hints do not prove an area is clear: a person may be obscured or too small.
3. Verify against the original
Open the crop and inspect it in the context of the full image. Compare neighbouring frames where possible. Set a status, add a note and reject obvious mistakes. The model’s displayed score is not the probability of finding the missing person. Refer suspicious locations for field verification under your team’s procedures.
4. Inspect the map
Use Show on map for a specific finding. Without complete telemetry the marker identifies the drone, not the person. A flat-terrain estimate needs altitude, camera angles and field of view; check its label and uncertainty radius. The searched-area layer records analysed imagery but does not guarantee detection of every person in that area.
5. Export and back up
Share CSV, KML or a report with the operation coordinator. Crop sharing uses your device’s sharing facility, not a Look for human upload server. Keep a .lfh.zip archive because clearing browser data can erase local history. Sector and team-track exports are separate from drone-footage exports.