Exporter vers disque permet de redimensionner en choisissant la résolution et les dimensions de l’image avec les algorithmes classiques Bicubique, Bicubique plus net et Bilinéaire. J’aimerai voir apparaitre un nouvel algorithme basée sur l’IA qui serait une référence dans le redimensionnement comme l’est DeepPRIME / DeepPRIME XD dans le débruitage.
I think he is asking for something like Gigapixel AI, resize with AI options, using similar style technology as DeepPrime, instead of old school algorithms like Bicubic interpolation etc. I support this idea, if DXO can divert the resources to it. I think there are options to resize images like said Gigapixel and others, but DXO showed that DeepPrime technology is top of the line, so maybe it can be used for up scaling as well. 2x up scaling trained AI would be very nice addition.
If you are making the image smaller, than you are throwing away pixels and traditional methods are pretty good for that. If you are upscaling you are interpolating new pixels, and you quickly find limitations with old methods, and that is where neural networks (AI) has the advantage. The only exceptions is vector graphics and things that has sharp edges. But in a continues tone raster image like a typical photograph, AI is the better solution.
I was just shooting some bugs with a 30mm macro lens, so I could not get close enough. Nice results from DXO, but it would be great to be able to upscale it on export 2X and get a better close up. As it is, I ended up using third party programs for up-scaling.