Search the Community
Showing results for tags 'artefacts'.
Found 1 result
We've seen a number of questions already arising about a "debug setting that miraculously improves texture quality". I'd like to explain the background and the underlying facts. Firstly though, let's establish a couple of facts. There is no magic button or debug setting to improve the resolution or quality of all textures. There is no way to display textures of a resolution greater than 1024x1024 in Second Life So what is all this muttering about and is there any substance to it? The "muttering" stems from some investigation by @Frenchbloke Vanmoer that was published by @Hamlet Au on his New World Notes blog with the title "How To Display Extremely High-Res Textures In SL's Firestorm Viewer" and in spite of the headline's conflict with the facts listed at the head of this post, yes there is substance to this news as it happens. I'll keep this post relatively short. If you want to see more rambling on how and why @Frenchbloke Vanmoer hit upon something interesting you can read about my subsequent investigation in my blog post, compression depression - tales of the unexpected. The bottom line is that whether by luck or by judgement the Second Life viewer uses a bilinear resampling algorithm when it resizes images. Until yesterday I, like many other, and I would suspect most of you reading this, had somewhat slavishly followed the generally accepted advice that bicubic resampling gave better results, more specifically that bicubic-sharper was the ultimate "best for reduction" choice. The evidence that Frenchbloke stumbled upon goes contrary to that advice and, in all my tests so far, for the purpose of texturing in SecondLife where you typically want to retain high contrast details bilinear gives better results. I should re-assert here, you do not need ANY debug setting. The original article used an obscure debug setting but it was only a means to an end, you are in general far better off and have far more flexibility if you use your photo tools as you always have. So what are bilinear and bicubic and why do we care? When you downsize an image, information (detail) gets discarded, deciding which information to keep and which to lose is behind these choices. All resampling methods try to decide which data to keep, or how to blend the data into some kind of average value that will please most people. Put simply a bilinear sample takes the 4 nearest points to the current pixel and produces a weighted average of those as the new value for the resulting output pixel. Bicubic takes this further, using 16 adjacent points to form its result. By virtue of the larger sample you get and smoother average which ultimately is why it fails us when we want to preserve details. On the flip-side of this is that for smooth gradients you may find more "banding" using bi-linear sampling. Why should we not use the debug setting? Firstly, as a general rule, debug settings are not a good thing to go playing with. They can frequently have side-effects that you do not realise and we often find that people tweak some random settings because "XYZ person recommended it" and perhaps it achieves their goal at that time, or as is often the case, it seems to fix things but doesn't really. In any case, they forget the changes and move on. A week or so later they are furious because things don't work anymore, they've forgotten all about the debug changes of course. More importantly in this case, if you use the max_dimension setting to force the viewer to rescale for you, then you will only see the benefit in 1024x1024 images. 1024x1024 is appropriate to large texture surfaces but not so much for smaller objects. If you can use a 512x512 you are using a quarter of the memory of a 1024x1024. That can make quite a difference to the performance of a scene. Many people remark that using a 1024 is the only way to get the detail that they feel that they need. I urge you all to take the lesson here as an opportunity to increase the clarity and sharpness of lower resolution textures by resizing from the large form originals directly to the target size in your photo tool of choice. Don't forget, and this may sound obvious, You need to have high-resolution images to start with. You cannot create something from nothing and whatever you do don't save the resized image to disk as JPEG before uploading, use TGA or PNG both of which or lossless. Give it a try today, and raise a glass to Frenchbloke while you marvel at the increased detail. A quick example My blog post above shows a worked example, but I thought I would show you another on a natural scene. Here, an original high-resolution image has been resampled down to an SL friendly 1024x1024 using both methods (entirely within photoshop to avoid all doubt around various other compression factors). First is the bilinear https://gyazo.com/545a21efb514ed16051f791ea9d527c4 Second I give you the bicubic https://gyazo.com/a969ec986746ced323e6f2f0ddbda0e8 On their own, they don't look that different, but the bilinear shows a lot more detail which is most noticeable in areas of high contrast such as the steps on the hillside