• Advertisement
Sign in to follow this  

Fast Box Filtering for Soft Shadows

This topic is 4143 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

Hi guys! I want to ask if anyone knows how to do the filtering and linear interpolation stated in this article Plausible Image Based Soft Shadows? Like how many pixel neighbours do we need to use? And how to linearly interpolate between 2 NBuffer levels and between 2 slices? thx! Edwin [Edited by - edwinnie on October 18, 2006 8:59:55 PM]

Share this post


Link to post
Share on other sites
Advertisement
no one knows?
or you cannot be bothered about this algorithm?

5 rating stars to anyone who knows!!

[Edited by - edwinnie on October 18, 2006 11:44:04 PM]

Share this post


Link to post
Share on other sites
Guest Anonymous Poster
Quote:
I want to ask if anyone knows how to do the filtering and linear interpolation stated in this article Plausible Image Based Soft Shadows?


Answer: Yes, someone knows how.

Quote:
Like how many pixel neighbours do we need to use? And how to linearly interpolate between 2 NBuffer levels and between 2 slices?


Question: Did you read and understand the article? I skimmed it but it seems to answer these questions.

Next question: What is "abt"? It's not a word in the English language as far as a few dictionaries are concerned. Then again you're in "Kiddy Grade" so that explains a lot about your bad English.

Share this post


Link to post
Share on other sites
what are you talking about?

Skimming through an article is NOT equivalent to actual understanding of an algorithm.

Share this post


Link to post
Share on other sites
Lets ignore the selfish and arrogant pple on board. Forums are supposed to be a shared resource. So I am sharing my thoughts.

Taking a quote from the article,
Quote:

D´ecoret introduced the NBuffers to allow prefiltering with continuously placed
kernels. We use them to compute the mean value of neighboring pixels. Each level l holds, for each texel, the normalized response of a box filter with a kernel size of 2^L x 2^L.


This "mean" value that needs to be computed is confusing enough. Texture lookup offsets for NBuffers are always north,east, and northeast of the current texel. Offsets are calculated based on power of 2 eg 1/256, 1/128 depending on the NBuffer level that is to be processed.

Would the "mean" be simply the summation of 4 texture lookups of the previous NBuffer level and divide by 4? What about the size of the filter kernel? Will or how does this filter size affect the normalization of the filter response?

Apart from these questions, notice the way NBuffers tries to do its texture lookups. Its always based on a north,east, and northeast direction. So far I am trying to compute an intermediate result similar to that shown in the video. The following image below shows a reference of the "intermediate results" which I am referring to:
Free Image Hosting at www.ImageShack.us

My current intermediate results look "skewed", and so I will ask another question: How to get those intermediate results shown above?

regards
Edwin

Share this post


Link to post
Share on other sites
update: with only 4 slices, this is what i get for inter-slice shadow intensity. Light leaks are not fixed yet and there is probably a few bugs.
Free Image Hosting at www.ImageShack.us

Share this post


Link to post
Share on other sites
I'm not convinced about the viability of taking planar "slices" in the long term. It seems to me that if you're trying to construct a representation of the depth distribution function it makes much more sense to do something like Deep Shadow Maps (i.e. construct a piecewise linear representation). Depth peeling seems to make a lot more sense to me in the long run - at least to one or two levels. Note that depth peeling + VSM will give exact, hardware-filtered results with no light bleeding. Furthermore depth peeling only N levels will eliminate light bleeding for the first N overalpping ocluder penumbrae... even with N=2 or 3 I suspect that light bleeding will effectively go away in most realistic scenes.

Anyways the one advantage of the slices proposed here is that they can be done without retransforming vertices. This is less interesting once we have StreamOut in DX10, but for now it may be worth it. I'm just concerned that it won't scale well to large scenes and may look bad with animated scenes (as objects cross between slice boundaries).

Share this post


Link to post
Share on other sites
I do not want to hijack the thread here.

Quote:
Note that depth peeling + VSM will give exact, hardware-filtered results with no light bleeding.

I would be interested in explanation and source :-)

Share this post


Link to post
Share on other sites
Quote:
Original post by wolf
I would be interested in explanation and source :-)

Check your PM - if others are interested in a discussion of this, we can start a new thread :)

Share this post


Link to post
Share on other sites


Might as well start a new discussion, as I am sure other people (including myself) would be interested. Would you be storing 2 depths and 2 depths squared, or just using the second depth as a reference depth ? Or is it more complicated than that (Not sure why you would need more than 2, but I'm no shadow expert ... yet :) )

Share this post


Link to post
Share on other sites
Quote:
Original post by multisample
Might as well start a new discussion, as I am sure other people (including myself) would be interested.

As requested...

Share this post


Link to post
Share on other sites
Sign in to follow this  

  • Advertisement