[digikam] [Bug 464266] Face detection is extremely slow - face recognition is almost impossible
https://bugs.kde.org/show_bug.cgi?id=464266 --- Comment #26 from Theliel --- The bug was opened due to speed problems when using facial detection. My comment was to point out that, at least with my equipment, as of today, the performance is adequate with about 2-3pps. I have not talked at any time about closing the bug, I have simply shown the results that I currently obtain For me at least, OpenCL works. It's not a dramatic improvement as long as it doesn't multiply the speed X times, but it works. Whether the performance is adequate or not would be a totally different debate. Obviously any improvement that can be made to it is welcome, and that is when Gilles himself encourages anyone who wants to collaborate and improve the current code. Obviously 404 photos is not an important sample, I have simply used it as a test. My current library is about 200k photos, there are currently about 150k people tags (about 600 different people), and another 80k detected and ignored. As you will understand, in my use, the speed to detect and identify is quite important. Greetings. -- You are receiving this mail because: You are watching all bug changes.
[digikam] [Bug 464266] Face detection is extremely slow - face recognition is almost impossible
https://bugs.kde.org/show_bug.cgi?id=464266 Theliel changed: What|Removed |Added CC||tahar...@gmail.com --- Comment #23 from Theliel --- Facial Recognition has evolved enormously, especially since the inclusion of Yolo-v3 (DNN barely recognizes me people, for me Yolo-v3 is mandatory). It is true that Yolo-v3 is slower than DNN, but currently the detection speed is more than acceptable. I just ran a couple of tests on a folder of 402 photos I had, almost all portraits and posed: (win11, 7700k, 4060ti) Yolo-v3: About 3pps (About 10,800 photos per hour) DNN: About 4pps Probably with CUDA it would be much faster, but I think they are acceptable times. On the other hand, if Apple wants to get rid of OpenCL to force developers to use Metal, it is a totally different matter, it is not precisely the fault of Digikam's dev. This is an open source project, if it were that easy to implement a new API, anyone could do it then... and that's obviously not the case. Currently as far as I know, OpenCV does not support Metal (and may not support it in the near future). The only way would be through Vulkan and MoltenVK, and it remains to be seen the performance in comparison. -- You are receiving this mail because: You are watching all bug changes.
[digikam] [Bug 450261] Digikam searches for sidecar files even though reading them is disabled
https://bugs.kde.org/show_bug.cgi?id=450261 --- Comment #3 from Theliel --- Thanks for the support. It seems to happen only when selecting an image, no matter if the properties panel is open or closed -- You are receiving this mail because: You are watching all bug changes.
[digikam] [Bug 450261] New: Digikam searches for sidecar files even though reading them is disabled
https://bugs.kde.org/show_bug.cgi?id=450261 Bug ID: 450261 Summary: Digikam searches for sidecar files even though reading them is disabled Product: digikam Version: 7.5.0 Platform: Microsoft Windows OS: Microsoft Windows Status: REPORTED Severity: minor Priority: NOR Component: Metadata-Sidecar Assignee: digikam-bugs-n...@kde.org Reporter: tahar...@gmail.com Target Milestone: --- I was debugging my PC's samba connections with Wireshark, when I started seeing constant xmp file access errors. I have never used sidecar files, but in any case I made sure that the read/write options on Digikam aboyt sidecar files were disabled STEPS TO REPRODUCE 1. Make sure that read/write sidecar files are disabled 2. Select any image in your library 3. Use any tool to see digikam file access. Observed for first time with wireshark 4. Digikam try to access to the xmp file sidecar (1.jpg, search for 1.xmp and 1.jpg.xmp in the same folder). OBSERVED RESULT Digikam try to find for each selected image, the sidecar xmp. even when the use of sidecar files is disabled. This is an avoidable waste of resources especially when the photos are being accessed remotely, for example via SMB EXPECTED RESULT Since the use of sidecar files is disabled, Digikam should never attempt to access sidecar files. It's not really a big problem, especially when we work with local files, but in remote files, where we usually have lower performance and higher latency, any improvement is welcome. SOFTWARE/OS VERSIONS Windows: 11 -- You are receiving this mail because: You are watching all bug changes.
[digikam] [Bug 440555] clear/Rebuilding face training data force all DB metadata to be written to images again
https://bugs.kde.org/show_bug.cgi?id=440555 --- Comment #4 from Theliel --- Created attachment 140483 --> https://bugs.kde.org/attachment.cgi?id=140483=edit Maintenance Window with options selected -- You are receiving this mail because: You are watching all bug changes.
[digikam] [Bug 440555] New: clear/Rebuilding face training data force all DB metadata to be written to images again
https://bugs.kde.org/show_bug.cgi?id=440555 Bug ID: 440555 Summary: clear/Rebuilding face training data force all DB metadata to be written to images again Product: digikam Version: 7.4.0 Platform: Microsoft Windows OS: Microsoft Windows Status: REPORTED Severity: normal Priority: NOR Component: Faces-Maintenance Assignee: digikam-bugs-n...@kde.org Reporter: tahar...@gmail.com Target Milestone: --- SUMMARY Every time the training data for facial identification is cleaned and rebuilt, DigiKam forces a resynchronization of all the metadata from the database to the images, in essence, the same result that would be obtained by selecting this option in the maintenance options. Work fine on 7.3.0 STEPS TO REPRODUCE 1. Go to Tool/Maintenance 2. Tick only Clean and Rebuild training data 3. Accept OBSERVED RESULT theoretically begins to clean and rebuild the database, but at the same time begins to rewrite all the metadata of the files from the DigiKam database too. EXPECTED RESULT It should clean and rebuild only the training data for facial identification SOFTWARE/OS VERSIONS Windows: Win10 21h1 -- You are receiving this mail because: You are watching all bug changes.
[digikam] [Bug 433716] New: Feature Request: Search/Detect upside down images through Deep Learning
https://bugs.kde.org/show_bug.cgi?id=433716 Bug ID: 433716 Summary: Feature Request: Search/Detect upside down images through Deep Learning Product: digikam Version: unspecified Platform: unspecified OS: All Status: REPORTED Severity: wishlist Priority: NOR Component: general Assignee: digikam-bugs-n...@kde.org Reporter: tahar...@gmail.com Target Milestone: --- SUMMARY It is amazing what can be achieved with Deep Learning these days. The improvement in face detection with YOLO has been incredible, and it has brought this idea back to me again. Apply Deep Learning to search / detect images that are upside down (or rotated), and thus be able to later select all of them (or those that one needs / prefers) to correct them. I have been seeing that there are many projects in this regard, and quite well documented. Perhaps some owned "sanitized" albums (with correctly oriented images) could be used as training data for the neural network, and later applied to the rest of the library. In my tests, this would also improve facial recognition (not detection). Those images that are already tagged and that are still rotated, greatly reduce the precision when used as training data for recognition. -- You are receiving this mail because: You are watching all bug changes.
[digikam] [Bug 416988] Objects / Forms / Monuments / Context detection and recognition using Deep Learning
https://bugs.kde.org/show_bug.cgi?id=416988 Theliel changed: What|Removed |Added CC||tahar...@gmail.com -- You are receiving this mail because: You are watching all bug changes.
[digikam] [Bug 371536] Enhancement: native 360-panosphere viewer support
https://bugs.kde.org/show_bug.cgi?id=371536 Theliel changed: What|Removed |Added CC||tahar...@gmail.com --- Comment #2 from Theliel --- I have to admit that until recently I didn't know about digicam (Windows user). I was looking for a good open source photo manager to manage, small photo editions, easy masive tagging... ... and with a decent facial recognition software. I have to say that I was pleasantly surprised, the only thing I miss for now is, effectively, a photosphere viewer. +1 for native 360ยบ viewer, maybe with different projections options -- You are receiving this mail because: You are watching all bug changes.