Choose a folder on this computer. Every JPG and PNG in it, including subfolders, is added to the collection. Other image types and photos over 5 MB are listed as not added.

The file name without its extension becomes the person ID, so LIC-1234567.jpg is stored as LIC-1234567. IDs can only use letters, numbers and _ . - :, so anything else, such as a space, is replaced with _.

People already in the collection are skipped, unless Replace is ticked. A photo with more than one face is added with a warning, so it can be checked.

Click a photo to see it full size, or to remove the person. Removing deletes their faces from the collection and their stored photo.

To remove several people at once, click Select, tick the photos (or Select all for everyone matching the search), then Delete selected.

People added by the script from SQL are listed as No stored photo. They can still be removed here.

Compares every face with the rest of the collection and lists pairs of different IDs that look like the same person, with the highest similarity found for each pair.

It runs one search for every face, so a large collection takes a while and has a small cost. Minimum similarity is the cut-off; lower it to catch more possible pairs, raise it to see only close matches.

When someone clicks an identity match on the dashboard, PhotoCheck sends the matched ID to its webhook and shows the photo that comes back beside the applicant. This server can answer that itself from the photos stored here.

On Elastic Beanstalk, http://localhost:8080/api/collection/webhook also works as the address: the server calls itself without leaving the machine.

People added by the script from SQL have no stored photo here, so the webhook answers "not found" for them.