digiKam 8.6.0 is released

Dear digiKam fans and users,

After four months of active maintenance and many weeks triaging bugs, the digiKam team is proud to present version 8.6.0 of its open source digital photo manager.

• New Features and Major Changes
  • Face Management Updates and Improvements
  • Auto-Tags Management
  • Image Quality Management
  • Red Eye Correction
• Internal Components Update
• Notable Bug Fixes
• Generalities
• Future Plans
• Final Words

The digiKam team has continued to work on a better Artificial Intelligence integration in digiKam, and many parts have been improved with the 8.6.0 release.

New Features and Major Changes

Face Management Updates and Improvements

digiKam’s Face Management tool utilizes cutting-edge AI technology to automatically detect and recognize faces in images. You can access this feature via the People tab in the left sidebar or through the Maintenance tool.

The face management framework in digiKam has been completely rewritten, including the Face Classifier and Face Pipeline implementations. These major modifications bring about significant enhancements in digiKam 8.6.0. The pipelines now operate 25%-50% faster when using the full CPU, drastically improving performance and efficiency. The face detector has been refined to reduce false positives, ensuring more accurate detections. Additionally, a brand-new face classifier (matching algorithm) has been introduced, leveraging cross-validating K Nearest Neighbor (KNN) and Support Vector Machine (SVM) classifiers. These advancements significantly boost matching accuracy, making the face recognition process more reliable and precise than ever before.

From the user interface, face management UI has been simplified. The older and less reliable SSD, YOLO, and OpenFace models have been removed. All processing is now handled by YuNet for face detection and SFace for feature extraction.

The method of loading and converting images to a format suitable for the OpenCV framework has been refined. Previously, in digiKam 8.5.0, images were converted to BGR format for processing. Images are now converted to RGB, since YuNet and SFace are optimized for RGB. This change has led to a measurable improvement in the accuracy of both face detection and face recognition.

Another key improvement is the optimized use of GPU processing in the pipelines. Images are transferred to the GPU for resizing and color format conversion. Additionally, the GPU-processed image is sent to the neural network layer, enabling the reference engine to process it on the GPU via the OpenCL framework. Alongside GPU performance enhancements, the face pipelines now utilize a dynamically scaling algorithm to boost parallel worker threads according to the user’s hardware and memory configuration. This allows digiKam to fully utilize the computer’s hardware capabilities.

We’ve also added a Face Image Quality Assessment (FIQA). FIQA removes small, pixelated, noisy, and blurry images from the training dataset used for the recognition model. This quick, multi-step process involves a Fast Fourier Transform filter for blur detection, and convolution and Gaussian filters for noise detection.

The final major update for face management is the face classifier that matches one face to another for recognition. Previously, in digiKam 8.5.0, a custom-made KNN classifier was used. In version 8.6.0, an ensemble approach has been adopted, utilizing cross-validating KNN and SVM internal classifiers from OpenCV, along with a single custom distance check. The new classifier is both faster and more accurate.

The only thing we’re really reusing from digiKam 8.5.0 are the YuNet and SFace models themselves.

Auto-Tags Management

The auto-tags feature in digiKam analyzes image content to detect various elements such as objects, statues, animals, plants, and events. Keywords are automatically generated by a neural network and can be translated into your preferred languages.

The Auto-tags engine has been completely rewritten, incorporating the new pipelines and improvements found in the Face Management engine. New classifiers and the option to adjust the Auto Tagging confidence threshold have been added. Auto Tagging is now faster and more accurate, utilizing the latest YOLOv11 Nano, YOLOv11 XLarge, and EfficientNet B7 Deep-Learning Neural Network models. To reduce the data file size at startup, the older YOLOv5 and ResNet-50 models have been removed.

The auto-tags scan tool is now available in the Tags tab on the left sidebar for improved usability. It remains accessible in the Maintenance tool and as a Batch Queue Manager plugin.

Image Quality Management

The Image Quality feature in digiKam assesses the aesthetic quality of images and automatically classifies them with a colored flag (pick label) in the digiKam database. The quality assessment can be determined by using modern AI engines or by simple algorithms (such as noise, blur, and exposure detection).

The Image Quality Sorter has been renamed to Image Quality Scanner and is now available from the Labels tab on the left sidebar for better usability. It remains accessible in the Maintenance tool and as a Batch Queue Manager plugin. The global Image Quality settings from the setup dialog have been removed.

Red Eye Correction

Both the Image Editor and Batch Queue Manager feature a plugin designed to detect eyes in a face and correct red-eye caused by the camera flash. These tools have been completely redesigned with a deep learning engine, delivering superior analysis and correction compared to previous versions of digiKam.

Internal Components Update

This version includes an update to the internal RAW decoder, Libraw, to the rolling-release snapshot from 2025-02-08, along with various bug fixes.

The industry-standard open-source component ExifTool has been updated to the latest 12.99 release in all bundles for metadata management.

All bundles have been updated to the Qt framework version 6.8.1

The famous G’MIC-Qt plugin available in Image Editor and the Batch Queue Manager is now updated to last 3.4.2. It’s now possible to apply a G’MIC-Qt filter on a selection square previously selected on the Image Editor canvas.

Notable Bug Fixes

• Fix wrong processing in the face management workflow 501257 498024 425263 425981 499301 415777 498104 495608 497791 498320
• Fix behavior in the face management user interface 389046 432760 499172 500140 463760 472416
• Fix crashes in the face workflow 500570 488237 500961 498603
• Fix video file metadata interoperability with proprietary softwares 498024
• Fix crash with WebM files under Windows 485468
• Fix Linux AppImage bundle desktop integration 493578 492813
• Add a new video decoding acceleration using GPU 484479 496484

We extend our gratitude to André Molkentin, our honorary Quality Test Engineer, for his countless hours of dedication in helping the team identify and resolve several issues with the Face Management workflow.

Generalities

140 bugs have been fixed, and we dedicated substantial time to reaching out to users to validate changes in pre-release versions. This ensured that fixes were confirmed before preparing the digiKam 8.6.0 release.

The internationalization of the application has been updated, and digiKam and Showfoto now support 61 languages for the graphical interface. To change the language, navigate to the Settings/Configure Languages dialog and select your preferred localization. Please restart digiKam to apply the changes. If you’re interested in contributing to the internationalization efforts, please contact the translator teams, following the translation how-to. The statistics about translation states are available here.

Thanks to the translators who have worked on the online documentation internationalisations which is available in 16 languages as French, German, Spanish, Italian, Japanese, Chinese, and more. You can read and search over the documentation here. You are welcome to contribute to application handbook translations following the coordination team instructions.

Future Plans

Next maintenance version is targeted for June 2025 with more bug fixes and improvements.

In the coming weeks, we will study more tools to use deep neural networks for automatic tasks, such as noise reduction, colors adjustments, rotation, etc. Another important feature is to see if a LLM engine can be used as a natural language interface to query the database for the search of items in the collections.

Final Words

Thanks to all users for your support and donations, and to all contributors, students, testers who allowed us to improve this release.

digiKam 8.6.0 can be downloaded from this repository as:

• Source code tarball.
• Linux 64 bits AppImage bundles compatible with systems based on glibc >= 2.31.
• Windows 10 (or later) 64 bits installers or bundle archives.
• macOS Silicon and Intel packages.

Rendez-vous in a few months for the digiKam 8.7.0 release.

Have fun with digiKam…