Bloated Model Images

[SneaksieDave]

At over 700Mb, the models directory is now the biggest size offender by a long shot. Some of that is necessary of course, but a lot of it is not. The Thief-style wall lamp is a good example, with the diffuse being a ginormous 4Mb (despite being blurry/pixelated), and for such a small object. There's no reason it can't use a 683kb DDS, by the same conventions that have already been accepted for geometry textures. There's no noticeable loss, and a space savings of around 75%.

 

That said, if no one objects, I'd like to take a side project (little bit here, little bit there, as the urge hits - that doesn't mean I promise to do it all, nor that outside help isn't welcome, but that's up to you) to convert them as appropriate to DDS. This would be diffuses and speculars only, as is also convention with the texturing division. It seems right now, the only model in the whole mod even using a DDS is the bell I made.

 

Objections?

[sparhawk]

Yes. As long as we haven't made sure that the DDS doesn't create additional problems, as it seems to, we should definitely not place all our bets on it. First we must find out what the problem with bonehoard is, before we can make such a move. If it turns out that it's not the DDS but something else, fine. Go ahead. If it turns out that this is the problem, then what do we do? Covnert everything back again?

[OrbWeaver]

Sounds like a good idea to me. Don't forget downsizing diffuse/specular maps as well, this alone will probably give you a better size reduction that using DDS (and combining the two is even better still).

[Springheel]

Yes, some specular maps are still much larger than they need to be. I resize the ones I come across to 128 (or 256 for really big models), because many of them are 512, which is completely unnecessary.

 

I'd double-check that D3 will automatically search out the .dds if it can't find the .tga, otherwise you'll have to change all the material files as well, which would be a major undertaking.

[SneaksieDave]

Okay so there is disagreement, and good points raised.

 

Anyone having trouble with bonehoard: are you able to see the bell properly (in props/misc)? I thought Angua had already determined it was an issue of using Ultra quality with the wrong texture compression setting? (admittedly I haven't followed it closely - I was never able to view DDS normalmaps, but speculars and diffuses are fine).

 

As far as the paths go, good point SH. If the materials are written wrong, for instance, pointing to *.tga instead of a filename with no extension as they should be, that could cause a problem. If written properly though, materials definitely reference DDS, because that's what the bell does.

[OrbWeaver]

I think Doom 3's handling of DDS is automatic -- you just point to the TGA as normal (with or without the extension, it shouldn't matter), and as long as there is a corresponding DDS file in an identical tree structure mirrored under the dds/ folder, it should be loaded automatically if compressed textures are in operation.

[SneaksieDave]

Just a quick caution as well, that speculars can't always just be blanket-sized down. Note the difference in the bell's specular, compressed to 256X256 alongside the original 1024X1024. I'd personally just go with DDS but retain the resolution.

[OrbWeaver]

Well it's obvious that that particular image won't respond well to downsizing, but the question is whether you will notice in the game. Provided that the diffuse and normalmaps retain sufficient detail, it should theoretically be possible to use a much lower-resolution specular map since this is only seen in certain circumstances and from certain angles.

[SneaksieDave]

So, are we any closer to finding out what - if any - texture problem there was floating around bonehoard? Is it new, or the known issue? Must it halt this, or is it unrelated?

 

I could do a conversion for just the wall light in question, as a test, and then those with bonehoard problems could check out that exact model to see if this has any relation...?

[ascottk]

I experimented with dds over the weekend on the builder priest & I used ATI's The Compressionator and the Gimp, & I used several settings from both programs. None worked for me. I tried mirroring the main model directory & without mirroring, dxt 3, 5 (rxg5b5 with ATI), uncompressed with several different settings & I only got black materials.

 

I used the recommended settings on mod wiki:

http://www.modwiki.net/wiki/Texturing

[SneaksieDave]

Also, I just tried dxt1c compression for DDS on the wall lamp as a test, and it turns out Doom3 doesn't accept that compression for alphas, I guess. Luckily, that texture has one, so it was a fortunate test subject. I even tried with the -alpha switch in the command line, and it didn't work. Someone clue me in if I'm doing something wrong. Does DDS work with alphas at all in D3?

 

So anyway, I cut the size of the image from 1024X1024 to 512X512 and that reduced the size by 75%, down to 1025kb instead of 4097kb. I did this all locally - nothing's being committed unless we can figure out the details. How is this problem still elusive? This should be a simple, documented, set of known limitations, somewhere on id's site.

 

There's no reason the "safe" resize-and-save route can't be taken. And ultimately no one's complained about the bell not working... then again, maybe people haven't checked (hint, hint - it would help this effort along).

[ascottk]

Also, I just tried dxt1c compression for DDS on the wall lamp as a test, and it turns out Doom3 doesn't accept that compression for alphas, I guess. Luckily, that texture has one, so it was a fortunate test subject. I even tried with the -alpha switch in the command line, and it didn't work. Someone clue me in if I'm doing something wrong. Does DDS work with alphas at all in D3?

 

So anyway, I cut the size of the image from 1024X1024 to 512X512 and that reduced the size by 75%, down to 1025kb instead of 4097kb. I did this all locally - nothing's being committed unless we can figure out the details. How is this problem still elusive? This should be a simple, documented, set of known limitations, somewhere on id's site.

 

There's no reason the "safe" resize-and-save route can't be taken. And ultimately no one's complained about the bell not working... then again, maybe people haven't checked (hint, hint - it would help this effort along).

DXT1 has no alpha channel while the DXT3 and above has alpha. The Compressionator does have options to use the alpha channel on DXT1 (1 bit) but I'm not sure that'll actually work.

[SneaksieDave]

Ah, wasn't aware of that. Crap, then that opens another whole can of worms - I thought to this point we'd gone with dxt1 because it was safest and most compatible. I could be wrong. But if not, and we go up to dxt3, that might squeeze another portion of users out of the loop.

 

Argh what a mess this all is.

 

Edit: Yep, dxt3 worked for the alpha. Here's the funny part - filesize: 1376kb. So the compressed dxt3 is 1/3 larger than the half res TGA (which was 1025kb). So even if we went with it, we'd have to decide: compressed and slightly bigger, or smaller, but half res.

[ascottk]

Ah, wasn't aware of that. Crap, then that opens another whole can of worms - I thought to this point we'd gone with dxt1 because it was safest and most compatible. I could be wrong. But if not, and we go up to dxt3, that might squeeze another portion of users out of the loop.

 

Argh what a mess this all is.

I hate working with dds files I did that a lot when working on TDS stuff. Also I found out there's a bug with nvidia cards rendering textures with DXT1 compression which might've been fixed recently.

[OrbWeaver]

Edit: Yep, dxt3 worked for the alpha. Here's the funny part - filesize: 1376kb. So the compressed dxt3 is 1/3 larger than the half res TGA (which was 1025kb). So even if we went with it, we'd have to decide: compressed and slightly bigger, or smaller, but half res.

 

That's because reducing the dimension size of an image decreases its total number of pixels by three-quarters. DDS compression does not achieve anything like that AFAIK.

[greebo]

@Ascottk: Which version of The Compressonator have you been using? There is only one specific version of the ATI tool that's working for Doom3. (You probably knew this, but I wanted to make sure.)

[ascottk]

@Ascottk: Which version of The Compressonator have you been using? There is only one specific version of the ATI tool that's working for Doom3. (You probably knew this, but I wanted to make sure.)

I didn't know that there was a Doom 3 specific one I have 1.30.1084.

 

EDIT: Found the download for it:

http://www.pcgamemods.com/mod/17588.html

[New Horizon]

I didn't know that there was a Doom 3 specific one I have 1.30.1084.

 

EDIT: Found the download for it:

http://www.pcgamemods.com/mod/17588.html

 

Check out this link. It outlines the steps iD used to create their DDS files. The steps are fairly specific, but if followed, the resulting DDS files should work as well as they did in D3.

 

http://www.quake3bits.com/htm/tutorials/cr...pressonator.htm

[greebo]

Should I upload this version to our FTP in case the links die or the 1.21 Compressonator becomes hard to obtain?

[sparhawk]

Yes. That would be a good move.

[greebo]

It's on FTP now: http://208.49.149.118/TheDarkMod/Files/ATI...nator_v1.21.zip

[Crispy]

I'd just like to point out that the non-downsampled textures should go in the hires repository on CVS, if there aren't corresponding higher-res textures there already. Don't permanently throw any detail away please.

[SneaksieDave]

Indeed. Surely there's a way to mass copy all TGA in low res over to the high res, right on the server? If so, might as well do it for every TGA in the whole mod - it'll help with both models and textures.

[Schatten]

Uh.. while we're on model textures: I was thinking about creating a few alternative outfits for the existing models. My question is: Will alternative skins be supported for AI models? I mean without overwriting the files of the existing ones.

[New Horizon]

Uh.. while we're on model textures: I was thinking about creating a few alternative outfits for the existing models. My question is: Will alternative skins be supported for AI models? I mean without overwriting the files of the existing ones.

 

Yup, you will be able to swap alternative skins.