![]() ![]() You just tag a bunch of people and it went off and 'did stuff' behind the scenes, after which you'd be presented with reasonably accurate suggestions. Picasa made this stuff simple circa 2009. When do I need to run "Tools > Maintenance > Detect and recognize faces"? What does "Face Accuracy" actually affect? Detection? Recognition? Both? except some suggestions aren't even faces. The only thing they have in common is that they 'have faces'. ![]() Lots of random faces that look nothing like them, or even like the other suggestions. ![]() The suggestions for each person are laughably bad. (Updated) Even below 70%, a blind cat with a dart board would be about as good. I have not yet tested ever 1% increment between 0% and 100%, but what I have found is that it goes from "trash" to "nothing". But what? How does the application give me so much scope to screw this up, without giving me a single clue as to what I'm screw up?ĮDIT: Clarification on my use of "Face Accuracy". It can't be this bad, or there'd be a lot of noise about it on forums. It has on occasion prompted me with a wildly inaccurate guess (a large bald man suggested for some girl). Aside from mostly prompting with 'the last name', it has literally never correctly suggested a person. For example, I correctly tag my sister (female, black hair, glasses) and then I click on her husband (male, ginger hair, balding, no glasses) and it prompts me with her name. ![]() I'd like to mention a weird behavior that someone else mentioned - when I start correcting the faces in "Unknown", it always prompts me with the last name I entered. I don't get any errors, but basically nothing shows up in "Unconfirmed". (I tried moving the Face accuracy from low to high, checked and unchecked YOLO v3)Īfter spending a few hours on it at this point, it hasn't managed to recognize a single person. (I did this bunch of times, with every conceivable combination of options) Go into Tools > Maintenance > Detect faces. All are tagged at least twice, but as the people get closer in my social circle, some (eg: my wife) are tagged over 100 times. First off, here's what I've done (on the latest 6.x version and the latest 7.x version):Īdd a few albums that contain a decent number of images of family and friends (just under 2000 photos) Digikam face metadata pdf#This allows the regions to be resistant to changes in size, as long as no cropping takes place, and rotation, as long as the Orientation of the image is properly updated.įor further info, see the MWG Guidelines for Handling Image Metadata pdf on the subject (pages 51-56).I'll head over to if it turns out that this is a bug - but honestly, I think I must be "doing it wrong" because the face recognition feature simply cannot be as useless as it seems to be. Also, it is assumed that the location is relative to the top left of the image as indicated by the Orientation tag embedded in the image. This means that they will correctly locate the face as long as the image has not been cropped or flipped. The way MWG regions (face tags) work is X/Y coordinates and the height/width of the region are saved as number from 0 to 1 and used as a percent of the images actual height/width. Add -overwrite_original to suppress the creation of backup files. If the names are different, then the commands would have to be altered a bit but it should still be doable. If the filenames are the same, then you could do it in batch with a command like this: exiftool -TagsFromFile /Path/to/Smalls/%F -RegionInfo /Path/to/Larges The basic command to copy would be: exiftool -TagsFromFile SmallFile -RegionInfo LargeImage As long as none of the smaller set of images have been cropped and only are resized, you should be able to do it with Exiftool.įrom my brief search, it looks like digikam saves the faces tags to the MWG XMP tags. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |