Archived

This topic is now archived and is closed to further replies.

ABT question

This topic is 5393 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

Recommended Posts

Hey, I''m doing a research on 3D engines and the journey has brought me to ABTs. I have a question about calculating the minimize functions Yann talked about here. I found this thread which helped me a bit by giving these four lines: Axis Score = Dimension(ThisAxis)/Dimension(LargestAxis) Volume Score = 2*fabs(0.5 - SplitPercent) Face Score = fabs(FrontBackDiff)/NumFaces Splits Score = fabs(NumSplit)/NumFaces As I can see the volume, face and splits score is the lowest when there is no difference in the volumes, when the faces are evenly put into the two new AABB-s and when the number of faces split is the lowest. Now my question is: why is the axis score calculated like that? As I understand it''s best to split the box along the axis which has the highest length. The axis score ''function'' above generates higher values for the axis with the highest length (as opposed to lower values like the other 3 functions calculate for the better case) and lower values for the axis with the lowest length. Since the final score of the splitting point and plane is the sum of all four, doesn''t this mess up the box-splitting? Or have I simply misunderstood something? Thanks in advance .
---
cone3d
http://cone3d.gamedev.net
My software never has any bugs - it just generates random featureshello?

Share on other sites
Yeah, it should really be
1 - Dimension(ThisAxis)/Dimension(LargestAxis)

-------------
andyc.org
Destiny3D

Share on other sites
Thanks a lot, I've managed to complete a chapter on ABTs. It can be found here (needs open-/staroffice, pdf version), if anyone should be interested. You'll find your name in the references part .

The only problem is that the entire document is in estonian . A translated version will come out some time after the research's due-date on the 1st of March.

---
cone3d
http://cone3d.gamedev.net
My software never has any bugs - it just generates random features

[edited by - cone3d on February 8, 2004 4:50:43 PM]

Share on other sites
lol .. staroffice ...
any real version coming out?

Share on other sites
I hope you create engilsh verison. I would like to see how other people selected weights for criteria function, or even chaneged it.

You should never let your fears become the boundaries of your dreams.

Share on other sites
quote:
Original post by Anonymous Poster
lol .. staroffice ...
any real version coming out?

lol.. estonian
Please translate it to english, for it seems pretty interesting, according to the few words I understand

I also find it great that you are going to spread the ABT further. Since Yann is gone, he obviously won''t spread it any more, and so afaik the only resource about it is the thread here on gamedev. I find it a great structure, and so it should live on.

My Site

Share on other sites
I''ve really enjoyed ABT''s (my own vesion here) But there''s still alot of issues with the process I''m trying to clear up. Namely very high poly, singular objects. In my current implimentation, a single polygon was split 17 times due to a 100k poly object that happened to lie on the split axis of the ABT node.

I''m interested in reading the research paper, English would be excellent!

~Main

==
Colt "MainRoach" McAnlis
Programmer

Share on other sites
An english version of the paper will come out some time after the 1st of March. That''s the date when the estonian version has to be complete or I''ll be in a lot of trouble.

The research paper will mainly talk about many different algorithms, but it will do so without any code. Code for some algorithms will be posted later when I create a 3d engine using some of the algorithms described.

1. 1
Rutin
27
2. 2
3. 3
4. 4
5. 5

• 11
• 9
• 9
• 9
• 14
• Forum Statistics

• Total Topics
633313
• Total Posts
3011316
• Who's Online (See full list)

There are no registered users currently online

×