Look for humanWhat the model reviews
D-FINE looks for objects in the person class. Large images are reviewed both as a whole and in overlapping regions; candidates are checked again in an enlarged crop. Processing takes place on the device. Capability and speed depend on browser support, memory and available acceleration, not on uploading the file to a server.
A result does not identify a person
A box flags an image region resembling a person. The model does not identify an individual, recognise faces or confirm that the subject is the missing person. Weak candidates are separated from detector-verified candidates. Operators independently set a review status to organise their work.
Where AI can fail
Shadows, trunks and rocks can resemble people, while small subjects, blur and vegetation can hide them. The thermal test so far showed weak detection and does not establish forest effectiveness. Treat thermal support as separate auxiliary heat hints, not reliable automatic person identification. Hints can be false alarms, and no hints do not mean no person.
How to use a hint responsibly
Inspect the original and close-up rather than relying on box colour. Compare video sequences, record the reason for your assessment and request a field check when appropriate. Map positioning needs telemetry: a box alone cannot locate a person. AI is an additional review aid, not a replacement for visual inspection or incident command.