Mikaeri said:
Though I do wonder because I haven't been paying attention all too closely, there are users that misappropriate image sample and downscaled? That's a little bit concerning.
Mostly the latter, I think. I've seen it tagged on images that are just smaller than their parents, but are e.g. from a different source and not (third-party) downscaled. And then there's this tagging of WIPs as samples that was mentioned in this thread. It's probably not as bad as I make it sound, but it just worries me a little that something legitimate might get flagged and buried.
CodeKyuubi said:
Does that include post #1966693? The last time a conversation was started on twitter pngs, they decided twitter pngs, though visually lossless, were inferior in quality to conventional pngs. Following that train of thought, even if the resolution is the same and the filetype is lossless, md5 not matching is md5 not matching, and isn't the one uploaded by the artist and therefore sampled and inferior.
I compared the actual pixel values for that image and the original from Twitter, and they're identical. There's no difference in metadata either; the only difference is that Twitter has recompressed it, losslessly, at a lower compression level. (Why they do that, I have no idea.) That is not the same as sampling. Although it would have been nicer to have the original, nothing would actually be lost here if the source disappeared. In other words, I don't worship the file the artist created, but the image. Preferring the artist's original file is a good general rule, because that makes things a lot simpler - it's a much easier distinction to make, and second-hand copies are lossy more often than they're not. Even with PNG being a lossless format, a sample could still have been altered before encoding. But in this particular case, after determining that the information that really mattered was there, I guess I figured it wasn't worth it to mess up favorites or whatever. All that said, if someone does want to upload the original, that post would be fair game for flagging, and I wouldn't have a problem with that. Maybe I'll even do it later, now that I've spent so much time on this anyway...
Mikaeri said:
I'm to agree with the comment above... If they are the exact same image pixel-by-pixel then so be it, but then one has to wonder what composes the rest of that file bulk. I've never tried it, but it seems a little contrary to believe that the sampled image might actually be better than the original image in that regard. But even then, it's been settled for some time now that :orig is the most original image available, regardless of whether :large is actually better.
Maybe I'll test it later if I have time.
It's not so much a matter of "rest of that file bulk", as in the larger image is the smaller one plus something. It's two different ways of storing the same thing, that take up different amounts of space. When you compress a PNG, you can spend more time in order to get better compression, resulting in a smaller file, or you can do it quickly and get a bigger file, because it isn't compressed as much. Twitter opted for the latter. But when you decompress the two files, you get exactly the same output. It's like if you put one file in a .zip and one in a .7z - those archive files will be different, but the file inside will be exactly the same when you take it back out of either one. You can even try a program like optipng, which spends a long time trying to find good parameters to compress a png to a smaller filesize - if you run it on each of the two files, it will make them identical (shrinking them both).
The :large is not "better" than the :orig, ever. Larger filesize does not automatically mean better. Take for example post #2553638, which is bigger in filesize than its parent but actually worse, because the second time it went through the jpeg encoder, it treated the artifacts left from the first time as data, and had to encode those too (which takes up space), but for the actual data the best it could do was the same as the original, and in some places it's worse. You can't do better than the original unless you guess at what should be there, at which point you're re-drawing rather than just processing. (Maybe waifu2x should be an artist tag...)
sweetpeɐ said:
I think it does. Let's say for instance a user found an image and could not trace down a proper source so they uploaded the best image they could find using a reverse image search engine. That wouldn't be an image sample but would be downscaled.
Here's why it matters. *_image_sample refers to images sampled from specific sources and we have documentation of how to the best of our knowledge these images are sampled and how to avoid using them and in turn replace them with the true source. Thumbnails aren't from such a source. To alias thumbnail to image sample would needless erase information about the images tagged as such and would muddy the meaning of image sample.
Okay, that's fair. You've convinced me. I was thinking of "image sample" technically, like a thumbnail is an image sample by definition, but our tag has a more specific meaning. I'm not sure if the distinction can be made reliably, but I do see the purpose of it.