Thank you very much!
I'll probably use separated images then.
I'm doing a sprite editor before start with the game engine itself. In that editor, a sprite sheet is loaded and then saved after a little process. So I did this question to decide the best format to save them.
With separated images the size and coordinates table isn't needed also.
So, for each sprite, I receive the sprite sheet, split it into multiple small images and save them all in a data file. This file will be loaded into memory by the game engine, and a vector of images are loaded from that allocated memory.
Edit: interesting, looking at setImage of sf::Sprite code I see that mySubRect isn't adjusted when an image was set before, I always have to call both setImage and setSubRect. Not big issue, but..