<html><body><div><div><span>Thanks Maik!</span></div><br><div><span>I did a zfs snapshot of my photo archive and gave it a whirl. It finished about 30 minutes ago.</span></div><div><span>Unfortunately it didn't have any noticeable effect.</span></div><br><div><span>At 90% I can tag faces in images that are 98% or greater similarity according to the duplicate finder and it won't recognize them.</span></div><br><div><span>If I drop it down to 80% it might get some of them....but it also gets some wildly inaccurate guesses....like it thinks a chocolate lab is my blond-haired, blue-eyed kid.</span></div><br><div><span>I'll go through everything again this weekend and see if I've missed some incorrectly tagged faces.</span></div><br><div><span>-A</span></div></div><br><div><div>On Fri Apr 7, 2023, 08:12 PM GMT, <a href="mailto:metzpinguin@gmail.com">Maik Qualmann</a> wrote:<br></div><blockquote style="margin:0 0 0 4pt;padding-left:4pt;border-left:1px solid #CCC"><div style="color:#212121;font-size:14px;font-weight:normal;line-height:20px">In the maintenance tool under face detection there is the possibility to <br>rebuild the training database. It will be rebuilt based on the confirmed faces. <br>Note that face recognition in digiKam only works with face rectangles that <br>have been automatically detected by digiKam. Face rectangles created by other <br>programs or drawn manually are recognized very poorly or not at all.<br><br>Maik<br><br>Am Freitag, 7. April 2023, 14:51:15 CEST schrieb Aaron de Bruyn:<blockquote>Hey everyone,<br><br>I've been playing with the face recognition in DigiKam for a few months now<br>trying to sort through about 500,000 photos, and I've run into a few<br>problems.<br><br>I'm making a few assumptions here, so please let me know if I'm way<br>off-base.<br><br>I started out by tagging a few hundred photos of various family members<br>(let's just say Person 'A', 'B', 'C', etc...). After I did that, I ran the<br>recognition at 90% and it managed to get a few images that were nearly<br>duplicates of the ones I tagged. I confirmed them.<br><br>Then I dropped it down to 80% and a bunch were detected incorrectly.<br><br>While going through several thousand images...at one point I accidentally<br>tagged 20 or 30 images of Person A as Person B.<br><br>I noticed there's no "Undo", and that's understandable based on my knowledge<br>of how training databases work. Anyways, I went in to Person B, found the<br>images of Person A and corrected them to be Person A.<br><br>But from that point on, I appear to be doomed.<br><br>The system still frequently detects Person A as Person B, and I've made the<br>mistake a few more times while classifying images.<br><br>Yesterday I went through every single face tag and corrected probably 200<br>images in total that were confirmed incorrectly.<br><br>Anyways, I re-ran the recognition tool, and it's still guessing people<br>incorrectly...<br><br>My guess is the training database "learned" my mistakes and correcting them<br>doesn't really undo the damage I initially did.<br><br>Is there a way to "reset" the facial recognition database and re-train it<br>from scratch by saying "The images I have already tagged as Person A,<br>Person B, etc... are correct"?<br><br>Is that what "Clear and rebuild all training data" does in Maintenance? I'm<br>hesitant to try that option as it will probably take a solid day or two to<br>run and I don't want to lose all the manual tagging I've done.<br><br>Thanks!<br><br>-A</blockquote><br><br><br><br></div></blockquote></div></body></html>