Machine learning on your own computer
Let Portman2 find the faces in your photos and what each one is of, with models that run on your own computer — no service, no upload, nothing leaves the house.
The Machine learning pane manages those models.
Capabilities
Each capability is downloaded and switched on by itself, with its size and licence shown:
- Faces — finds faces, to place a photo's anchor: the part that thumbnail crops and Ken Burns pans keep in view.
- Salient subject — finds what a photo is of when there is no face: the anchor's second choice.
- Keywords — words for what is in a photo, so
Ctrl+Ffinds it by typing.
Use the switch to turn one on, Download to fetch it (one at a time), and Remove to delete it and switch it off. A capability switched on but not downloaded does nothing until it is. What they find can always be worked out again, and a rebuild throws it away.
Keywords for search
- Model — which keyword model the library uses, with its size and whether this computer has a usable GPU. Switching asks first: every computer then downloads the new model and analyses the library again.
- Cut-off — how sure the model must be before a word counts. Lower finds more and is wrong more often. Reset to puts back the model's default.
- Edit keywords… opens the keyword editor and Refine cut-off… opens Refine the cut-off.
The model, the word list and the cut-offs belong to the library and are shared with its other computers.
Analysing
Analyse runs the switched-on models over the photos already in the library, with Stop. It can be stopped and resumed.
Analyse new photographs automatically runs them on new photos by themselves once the library is idle; it steps aside for a scan or an import and waits while the computer is on battery.
The last line says where the models are kept: outside the install folder, so they survive an upgrade.