I’ve been doing some experiments on a post-process AO effect, based on this very interesting paper: Image Enhancement by Unsharp Masking the Depth Buffer.
I’ve got some pretty good results now, just using standard nodes in 2.43, and wrote up an article about it on my blog. I won’t re-post it all here, so click on the image to read the post. Hope you find it interesting!
wow very nice study, thanks a lot for sharing it with us, i love AO, and i think i will use it more now faking it, not abusing on render times is always a good idea :D.
Glad you found it interesting! 
yow! That is a nice shadow. Especially on the shirt around the tie and the jacket. And that is exactly what I was looking for too. I was planning to bake a shadow down and add textures to avoid a 12 hour AO render but now. . . . Hope this’ll work.
by the way, the .blend seems to be missing. Clicking the link delivers an error. Could probably use that here.
Oops, sorry about that, fixed now!
Well, I managed to duplicate your configuration. I applied compositing input from a large scene (well large for me about 42 blender units) so I don’t know if that made a difference. There was some color change in the preview but, as it turns out, difficult to say if the rendered image would have shown the subtleties that the underground man displayed as I also connected that output to a defocus node and blender dropped my computer to it’s knees. One thing that I couldn’t get was the outline like preview after the “tweak the mask strength across different areas”. The preview images was very solid. I also used a Render Layer node image as input instead of an exr image so maybe that made a difference too.
Enough for today. I was kind of hoping to have a fast shadow like that underground man has without having to bake, but with the hammering this PC took after connecting the defocus node makes me think that isn’t going to play so well.
Maybe I’m stating the obvious, but you didnt just re-use the settings though did you? You have to adjust the size of the first Map value node so it’s outputting a smooth gradient from white to black for the entire depth of the image. You may also need to try different values in the math/multiply node to scale it differently.
I’d just be curious anyway to see if this technique is dependent on certain kinds of scenes. Perhaps if you wouldn’t mind sending me an EXR of that render with colour and Z channels, I’d love to have a play.
cheers
Well, no, not so obvious. I did adjust the distances back and forth and could see the models display go from black to white. Also inverted the curve to see what would happen and that was, of course, incorrect. Knowing the distances have to cover the depth of the scene is a great help. I wonder if there’s any way to translate that to numbers. I was also wondering if you’ve tried the method on a horizontal shadow. Since the z value is what is generating the effect it brings up the question of whether or not the technique will work on surfaces not parallel to the camera plane.
My node setup was lost when the process was killed due to lack of horse power and I’ll need to rebuild. At least now I’ve gained some more experience and that is much appreciated.
btw it would appear the archive you’ve posted to your website is corrupt or something. Four different zip utilities all fail to open the zip so can’t do any science with that, my apologies.
At any rate, kind busy here with that contest thing Curtis has going over there. So scienctific pursuits are going to have wait for a while. Maybe next weekend.
Hi, I’ve fixed the .blend file, sorry about that!
Hey, Matt thanks for this, still have to experiment with this setup of yours, see what every piece really does. I was the one with the trouble finding the math node, was using 2.43RC1, dloaded the 3 and found it there. Tested it on a scene, scene from a challenge at cgtalk. You asked to see if it can be used on other scenes, so here’s mine…One without your method, the other with. It’s not really a good example of using your technique but the difference shows, at least a bit…Thanks again for sharing…Cheers
Ah, I see! It’s a very subtle difference, but it is there. Perhaps you could try fiddling with the multiply node to make it darker, and/or the size of the blur.
cheers!
I did as you said Matt, and played around with other settings, added a sharpen node, thought it made it a little bit stronger, darker, changed the multiplier and the blur too. I don’t know if it’s because I don’t know how to use this properly but I don’t think you can use this everywhere, like in this scene, you get the shadow behind that pole in the center, looking like the back wall is right behind it. I probably didn’t set something ok right? And I was wondering if there’s a way to ‘mask the mask’ so the blur from it doesn’t spread inside of objects, instead of just appearing behind it? I tried doing it in some ways but couldn’t do it…I’d probably use it to mask out the pole too, as it looks weird…Probably just need to make a separate mask right? not using nodes…? Much appreciate any assistance with this, really wanna see how best to use it.
Yes, I think you can do this if you use a Difference mix for the unsharp mask instead of the Subtract. In the paper, they use difference, and if I remember correctly it generated both light areas and dark areas, light for in front and dark for behind. But doing it this way will probably involve quite a bit of changing/tweaking to the rest of the setup, since what I have there is pretty tailored to that specific image.
I’d recommend reading the paper through to really understand what’s going on - it’s really not that technical, and only a few pages.
But sure, I bet it won’t work beautifully for all images, but if it works for most, then it’s not doing too bad 
I’m gonna listen to you and try to decipher that paper, tried reading it already but kinda got lost at the beginning…gotta try again with a little more concentration :yes: I’ll see if I understand anything and rather it can be used with difference but I doubt I’ll come up with anything better than what you’ve done…I agree with what you’ve said:
But sure, I bet it won’t work beautifully for all images, but if it works for most, then it’s not doing too bad
I mean I don’t think there’s a technique that can be used on every single scene…
