Availability
The version is already available for download on Windows, Linux and macOS. If automatic updates are enabled, the update will be installed automatically on Windows and macOS.
Function of forced assignment of persons on selected photos
Up to this point, all the tools for automating mass work with faces were designed for a more or less passive position of the user. The program itself suggests adding new persons, itself suggests new faces for the added persons. The user reacts to the program’s suggestions, evaluates the suggestions and accepts or rejects them. The suggestions come in fairly large batches, which makes the process very efficient in general - the user’s actions are required to be minimal, and the speed of face processing is very high.
At the same time, from communication with users it became clear that not everyone is comfortable with this mode. Some users want more control over which folders the program processes, want the ability to focus on specific photos and events right now, without waiting for the program to decide that it is time to start offering faces from there.
The program has all the necessary tools for such work too, but the efficiency (number of manual operations per face) is significantly lower, since you have to click on each face with the mouse. And when there are tens of thousands of photos, efficiency is very important.
After much thought over various options for solving this problem (several options were considered, including those that users suggested on the forum), it was decided to make one extremely simple function as a first step, but which, nevertheless, would speed up the process by tens and possibly even hundreds of times in the manual photo processing scenario.
In version 1.12, two hotkeys were added: Control+J and Control+Alt+J .
Control+J works in almost any mode, and can be applied to both the current photo and a large number of selected photos. This command forces the program to assign the most probable person (the one you will see first in the context menu as the “This is XXXX” command) to all faces that have not yet been assigned persons, if the faces and person have a high degree of similarity. If there is not enough similarity, nothing will happen.
Control+Alt+J does the same, but more aggressively - the threshold of similarity for triggering is much lower. But because this option often produces errors, the scope of the function is limited to the photo viewing mode so that you can immediately notice and correct them.
Please note that faces assigned in this way affect further recognition, and that errors are quite possible even in a conservative scenario (Control+J), so it is necessary to check the results visually . Otherwise, errors will only accumulate, invariably worsening the quality of all subsequent triggers, until the results become complete garbage.
According to our measurements, the conservative scenario gives no more than 2% of errors, normally processes most faces that look at the camera, but avoids faces of people looking away. The aggressive mode also tries to process all faces, but does not always guess correctly (but still guesses right more often).
Thus, it is now possible to scroll through photos one by one, press Control+J, Control+Alt+J, Control+I and only occasionally correct mistakes with the mouse.
We’d love to hear your feedback on this new photo workflow. Do you like it or not? Does it really speed up your work? What features have you noticed? How do you like the response thresholds for both options? How common are the errors? Please share your opinion.
