GUIDE [WIP] DDSopt & Texture Overhauls

[z929669]

Discussion thread:
DDSopt Guide by STEP
Wiki Link

---

 

GET DDSopt:
Github Pre-release versions
Official Nexus versions (select pre-release update 4)

[z929669]

You might want to simply verify those three textures. I am wondering if either I ran my input previously using alternate settings or if Vano himself made minor changes without changing the file name... sort of a patch without revision update? Note that I also have 20 fixes (I obtained my file from Vano a month or more ago)

 

I am pretty sure that I did nothing wierd, but I won't count on anything until we have the answer (it is mine or Vano's issue, as DDSopt will do what it does regardless)

[stoppingby4now]

As for "If you are lazy and have concerns about your VRAM' date=' you can skip to step 8 below"' date=' I don't think it's helpful to suggest people who'd rather download one of the HRDLC Optimised packs, rather than go through all the hassle of following the steps manually when no specific or significant benefit is identified are lazy (and indeed using HRDLC Optimised seems the sensible choice in terms of reducing VRAM usage).[/quote'']Ahh, but here is the real kicker: DDSopt actually IMPROVES some of the vanilla textures. I still have to verify IF these kinds of corrections are apparent in Vano89's Optimized (evidence of DDSopt), but for ex. fir trees in vanilla are wrong, and this is corrected by DDSopt'imization. This is a result of the mipmap adjustments that DDSopt makes to images as they are reduced. You can see the differences:Vanilla HRDLC before & after DDSopt (load both images and switch back & forth quickly to see the rather large difference. This is not apparent at close-up foliage and more pronounced as foliage distance view increases).

Likewise, do the same with these before / after images (look at the ground as you switch back & forth)

I have to disagree with the trees. When you are up close to the tree, it is full looking. The non-DDSopt trees in the background look as they should, fuller. The DDSopt version takes detail away. I don't see how that is correcting anything. I'd be more inclined to believe that those images were actually swapped.

[z929669]

 

As for "If you are lazy and have concerns about your VRAM' date=' you can skip to step 8 below"' date=' I don't think it's helpful to suggest people who'd rather download one of the HRDLC Optimised packs, rather than go through all the hassle of following the steps manually when no specific or significant benefit is identified are lazy (and indeed using HRDLC Optimised seems the sensible choice in terms of reducing VRAM usage).[/quote'']Ahh, but here is the real kicker: DDSopt actually IMPROVES some of the vanilla textures. I still have to verify IF these kinds of corrections are apparent in Vano89's Optimized (evidence of DDSopt), but for ex. fir trees in vanilla are wrong, and this is corrected by DDSopt'imization. This is a result of the mipmap adjustments that DDSopt makes to images as they are reduced. You can see the differences:Vanilla HRDLC before & after DDSopt (load both images and switch back & forth quickly to see the rather large difference. This is not apparent at close-up foliage and more pronounced as foliage distance view increases).Likewise, do the same with these before / after images (look at the ground as you switch back & forth)

 

I have to disagree with the trees. When you are up close to the tree, it is full looking. The non-DDSopt trees in the background look as they should, fuller. The DDSopt version takes detail away. I don't see how that is correcting anything. I'd be more inclined to believe that those images were actually swapped.

I am sticking to my guns on this one. Only way to tell for sure is to compare the details close up from the far away images. I believe that the "lost detail" you speak of is really just a distortion caused by rescaling the mipmaps. This is why you do not see this "bushiness" close up. I think that we should go to the experts on this & present to Ethatron for input... Correction or loss?

[z929669]

I invited Ethatron et al. to comment on this topic (from DDSopt thread on the Nexus)... let's find out for sure, eh? I'd rather add fact to the guide than fiction, and I have not gone into this as yet.

[Vond]

You might want to simply verify those three textures. I am wondering if either I ran my input previously using alternate settings or if Vano himself made minor changes without changing the file name... sort of a patch without revision update? Note that I also have 20 fixes (I obtained my file from Vano a month or more ago)

 

I am pretty sure that I did nothing wierd, but I won't count on anything until we have the answer (it is mine or Vano's issue, as DDSopt will do what it does regardless)

 

 

Finally managed to get on Nexus again (certain sites have not worked for me at all tonight, ISP issues I suppose) and re-downloaded. Noticed differences right away from my last run on them with DDSOPT. I most likely had already done some sort of optimization on the files before, or something was just off. :)

[stoppingby4now]

The tree in the immediate foreground has the same general fullness between the two. I'm mainly referring to the trees in the background. If those images are correct, I just don't agree that the DDSopt versions look better. Particularly when looking at some of the trees on the right hand side, there is very noticeable loss of detail. Which is why I was wondering if the images might have been labeled wrong (have you verified that you didn't make any ini changes that might affect the scene?). If not, I can't agree with DDSOpt version in this instance.

[z929669]

The tree in the immediate foreground has the same general fullness between the two. I'm mainly referring to the trees in the background. If those images are correct, I just don't agree that the DDSopt versions look better. Particularly when looking at some of the trees on the right hand side, there is very noticeable loss of detail. Which is why I was wondering if the images might have been labeled wrong (have you verified that you didn't make any ini changes that might affect the scene?). If not, I can't agree with DDSOpt version in this instance.

 

 

I have verified that this is pure DDSopt effect. I like the DDSopt better, and think that this is more true to the tree's close up form. Not a loss of detail between full-size and distant LOD in my opinion, but more a loss of added detail that does not exist close up.

 

 

 

We may be in the realm of aesthetics here, but I will see about blowing up some of those distant trees or creating a better compare.

[stoppingby4now]

That is something we'll just disagree on. I'm talking about distance trees. You may just like the look of the DDSopt version, but to be honest, it's not realistic. You don't lose that much definition and detail when looking at trees in that distance range in real life, and there is a lot of loss in both in the DDSopt version. I'm going to have to clear out my mods and put back the DDSopt I did to compare. Again, I can't help but wonder if the images were labeled wrong and we are comparing the opposite image.

[torminater]

Is there any reason for processing huge amounts of textures multiple times? How do I do that with DDSopt, can it cause problems, does it mess up files from user-created mods? Do I have to unpack .bsas before processing them?

Would love to hear some experts on that matter! :)

[frihyland]

Is there any reason for processing huge amounts of textures multiple times? How do I do that with DDSopt, can it cause problems, does it mess up files from user-created mods? Do I have to unpack .bsas before processing them?

Would love to hear some experts on that matter! :)

 

 

DDSopt will mark a file that it has already processed and will not reprocess that file unless you change parameters. You can process a bsa without unpacking it and repack it into a new bsa all with one click (and a few hours wait time).

[z929669]

A more detailed analysis, and images that can be saved locally and blown up for better scrutiny... smaller images are png format (no anti alias), larger images are jpg. After scrutinizing at this level, I have to admit that s4n has a point. I wtill argue that actual detail is not lost, but certainly information is sacrificed. Near-mid-distant trees arguably look better in DDSopt'ed, but distand trees are better in the unoptimized textures. Overall, I concede to s4n :facepalm: :P

 

 

vanilla

DDSopt

DDSopt(L)-vanilla®

vanilla

DDSopt

DDSopt(L)-vanilla®

vanilla

DDSopt

vanilla

DDSopt

[torminater]

So is it recommended to process them more than once? where do I change those parametres?

[frihyland]

There is the potential to lose quality each time you process depending on your parameters, that is why a flag is set so you don't process more than once. The parameters are covered in the ddsopt guide.

[z929669]

So is it recommended to process them more than once? where do I change those parametres?

 

 

Processing more than once, even forcing, will cause no change if the constraints and other parameters are identical to previous run(s). There is no point in doing this though.

 

 

 

@anyone

Do we have a tool like refscope available yet so that I can identify specific textures in-game? Given that fir foliage mipmaps are adversely affected to varying degrees, I'd like to identify the affected textures and see about either a different treatment or exclusin entirely.

[stoppingby4now]

A more detailed analysis, and images that can be saved locally and blown up for better scrutiny... smaller images are png format (no anti alias), larger images are jpg. After scrutinizing at this level, I have to admit that s4n has a point. I wtill argue that actual detail is not lost, but certainly information is sacrificed. Near-mid-distant trees arguably look better in DDSopt'ed, but distand trees are better in the unoptimized textures. Overall, I concede to s4n

 

There is information (definition) loss, but I would still argue that it equates to detail loss. You can't detail an object nearly as well with less of it there to work with. If information is being lost via DDSopt, logic dictates that some level of detail is going to go with it. I'm no expert on texture compression, and I doubt I ever will be, but basing on the constraint settings where they are all lossy, doesn't that mean that some detail is going to be sacrificed?