Chris Please Read This!

[selden]

Chris,

Unfortunately for that hypothesis, the image file contains no alpha channel, and the material's opacity is set to 1.

However (!!!) my billboard object has the default Anim8or values for all the material parameters, which includes a "diffuse" value of 0.7.

And another However: I just recreated the model, with diffuse set to 1, and it makes no difference.

I guess I'll have to try creating a CMOD billboard and see what that does.

*sigh*

This evening.

[selden]

I found the problem.

Anim8or includes an RGB parameter in the definition of materials which has the default value of

Code: Select all

rgb { 224 224 224 }

Editing the .AN8 file and setting it to

Code: Select all

rgb { 255 255 255 }

, reloading that model file into Anim8or and exporting it as a 3DS model eliminated the spikes.

I'm tempted to post a furious nastygram to the Anim8or forum, but I doubt it would do any good. It would seem to be one of those "Of course you have to..." things that a 3D novice like myself would never think of.

*sigh*

[jim]

I found the problem.

Nice to hear that this problem is now solved and my prepared post is no longer necessary.

Selden can you now make a test that compares the rendering of dxt1c and dxt3 textures (no alpha used) in Celestia. I think this could be very interesting. Maybe somewhere with an ATI card should also make this test.

Bye Jens

[selden]

Jens,

Here are the histograms for the updated model. Don't forget that the dxt1c texture created by nvdxt is somewhat different from the dxt3 texture.




The u888 and png and the dxt5 and dxt3 images seem to have identical color distributions, as they should, while jpeg and dxt1c are degraded.

[maxim]

That looks quite more reasonable, doesn't it?

maxim

[jim]

Don't forget that the dxt1c texture created by nvdxt is somewhat different from the dxt3 texture.

Selden, you are right! You have discovered a little bug of nvdxt.
Now I checked that nvdxt produces only good dxt1c textures with this switch: -24 dxt1c or -32 dxt1c. If "-dxt1c" is used nvdxt use the compression method of "-dxt1a" which cause less colors. For all the other formats (dxt1a, dxt3, dxt5) has the color depth switch no effect. Further I warn to use the -quick option this produce the lowest quality.

I made yesterday a lot of test with dxt1 compression and different tools and also I do now better unterstand Nvidea's dxt1 bug. I will post my results soon in the texture forum. But at the moment I can say that this bug causes that all dxt1 textures are only rendered with 32768 colors (15bit) instead of 1678840 colors. Believe me I was surprised that dxt compression allows such a high color number.

Jens

[t00fri]

Don't forget that the dxt1c texture created by nvdxt is somewhat different from the dxt3 texture.

Selden, you are right! You have discovered a little bug of nvdxt.
Now I checked that nvdxt produces only good dxt1c textures with this switch: -24 dxt1c or -32 dxt1c. If "-dxt1c" is used nvdxt use the compression method of "-dxt1a" which cause less colors. For all the other formats (dxt1a, dxt3, dxt5) has the color depth switch no effect. Further I warn to use the -quick option this produce the lowest quality.

I made yesterday a lot of test with dxt1 compression and different tools and also I do now better unterstand Nvidea's dxt1 bug. I will post my results soon in the texture forum. But at the moment I can say that this bug causes that all dxt1 textures are only rendered with 32768 colors (15bit) instead of 1678840 colors. Believe me I was surprised that dxt compression allows such a high color number.

Jens

Oh, I had not followed that thread...But I knew that
one has to take -24 dxt1c etc. Too bad..

I also agree on the low quality of the -quick option! It is certainly much worse than my DeVIL-based code in my texconvert tool

(see my discussion in the texture dept.

Bye Fridger

[Guest]

Now I checked that nvdxt produces only good dxt1c textures with this switch: -24 dxt1c or -32 dxt1c.

Foreget this! There is really a bug in nvdxt with dxt1c compression. The switch "-24 dxt1c" or "-32 dxt1c" cause only that the dxt3 is used instead of dxt1c. It seems that only the Photoshop plugin and Pixel's "bmp2dxt" produce good dxt1 files.

Bye Jens