https://bugs.kde.org/show_bug.cgi?id=518410
--- Comment #1 from Ondrej Zizka <[email protected]> --- Technical Implementation: For MariaDB (11.7+), digiKam can utilize native VECTOR types and HNSW indexes. For SQLite, I suggest bundling sqlite-vec, a lightweight, C-based extension that provides extremely fast vector search without external dependencies. This ensures that the heavy lifting of vector similarity is offloaded to the database layer, which is optimized for these operations, rather than rebuilding a FLANN index in application memory. Efficiency: HNSW organizes vectors into a multi-layered graph, allowing for incremental updates. Unlike FLANN, which often requires a full re-balance or rebuild to stay efficient, HNSW allows for fast insertion of new face vectors. This will reduce face confirmation times from several seconds to milliseconds, even on massive databases. Benefit: Users will experience near-instant face confirmations and significantly faster "Recognize Faces" passes. By adopting HNSW and sqlite-vec, digiKam remains a professional-grade tool capable of handling professional-sized archives. This modernizes the architecture to be "Vector-Native", significantly improving the long-term maintainability of the FacesEngine. -- You are receiving this mail because: You are watching all bug changes.
