https://bugs.kde.org/show_bug.cgi?id=478426
--- Comment #21 from Niklāvs Koļesņikovs <89q1r1...@relay.firefox.com> --- Above all else, I just do not see the use case for computer vision here. The requirement is for a nice scaler function, so there's plenty of choices available: Lanczos (sharp), Gaussian (smooth), spline16 (comparatively fast and for a lack of better term neutral). The list goes on but traditional scalers consist of either a single or multiple convolutions and often can be optimized into a single function that might fit on a single line of code. There should be suck optimized functions in https://code.videolan.org/videolan/libplacebo/-/blob/master/src/filters.c?ref_type=heads that probably can be copy, pasted and modified with a copyright notice, since ultimately it's just math. OpenCV is convenient but it is also weird in using the BGR format and depending on distro choices OpenCV packages may contain externally contributed code. On my system it also pulls in half a dozen new dependencies, including the same LAPACK that CImg optionally uses. Requiring something as large as OpenCV just for a scaler function strikes me as suboptimal. And the Tesseract OCR can be integrated directly on the account of it being C++. -- You are receiving this mail because: You are watching all bug changes.
