https://bugs.kde.org/show_bug.cgi?id=441092
Simon Gaschler <[email protected]> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |[email protected] --- Comment #3 from Simon Gaschler <[email protected]> --- I would like to add a current real-world use case for this feature request. I am using digiKam 9.2.0 on Windows with a collection of more than 200,000 images and a large number of confirmed faces. Rebuilding all face training data can take many hours. Running the complete rebuild as one uninterrupted maintenance operation is therefore impractical and fragile. Although training data is normally updated whenever faces are confirmed, a complete rebuild is still important in situations such as: - migration to a new installation or database, - loss or corruption of the training database, - interruption or crash during rebuilding, - changes or fixes in the face-recognition engine, - checking whether all confirmed faces have actually been included. For my collection, I have successfully tested an external batch-based rebuilding tool. It processes a configurable number of confirmed faces, stores its progress persistently, and continues from the last checkpoint on the next run. For example, I can rebuild 50,000 training entries at a time instead of requiring one uninterrupted run. It also provides an explicit option to reset the progress and start again. This demonstrates that a resumable workflow is both useful and technically feasible. Ideally, digiKam's maintenance tool would provide: 1. persistent progress/checkpoints, 2. automatic continuation after cancellation or restart, 3. an optional batch limit, 4. an explicit “restart from the beginning” or reset action, 5. clear information about whether the rebuild is complete or incomplete. For large collections, this is not merely a convenience. It is necessary to make recovery and maintenance of the face training data reliable. -- You are receiving this mail because: You are watching all bug changes.
