We recently gave a try at Adapive images on one of our testing Joomla website in order to solve the problem of serving scaled images and support retina. The result we are getting with it actually are strange, images being effectively resized but in most cases not at the right size, or at least not at the one we got here :
for the viewports of the devices tested.
Do you think it's a problem we would also have with XT adaptive images or is it really different than manually implementing Adaptive images ?
And if it's really different, could you please explain us what's the difference.
I understand perfectly that it can't match the exact size of the rendered box but actually it does not either match the viewport size.
For example on my desktop PC which has a viewport of 1536 x 710, it serves images of 1280 px width, meaning that those being displayed full screen on the website have to be scaled up by CSS to reach the viewport, increasing of course the blur effect.
I also noticed that the sharpen function completely change images colors instead of just sharpening them so we can't even use it to help on that point.
If you have a viewport of 1536 x 710, then images have to be of 1280 px width to support the case of a rotation. (The browser does not reload a page when it is rotated, and it displays the images served originally in a different page width).
We have no issues (or received reports) about sharpening issues. The Sharpen function (imageconvolution) relies on your GD library implementation of your server. So, your issue could be caused by a local configuration. You can disable the Sharpen feature in the configuration settings.
I do not really understand the point in resizing images lower in order to support rotation. Correct CSS rules should be able to care for rotation and images should be in my opinion resized to the biggest possible width of the viewport in order to avoid them being scaled up by CSS if used in full screen.
And if rotation is really the reason of the lower size, it should then be 710 px large in my example not 1280 px that is larger than the rotated device width, no ?
Anyway your answers brings a new point of view as I was thinking the size problem was coming from a bug or an incorrect guess of the viewport size. I really did not think it was wanted.
Are you absolutely sure it's a normal behaviour and if it is, do you know how the size is calculated.
I noticed during my tests that for a viewport of 1536 x 710, images were 1280 px wide and for a viewport of 1884 x 802, they wers 1536 px wide. I do not find the rule they follow