Percent Closer Soft Shadows by spidersharma03 #8933

bhouston · 2016-05-19T17:13:41Z

I'm submitting this PR but it was written fully by @spidersharma03.

This implements Percent Closer Soft Shadows (PCSS) as discussed here in issue #7653

The results look like this:

bhouston · 2016-05-19T17:15:33Z

This is a rough implementation, it likely needs to be polished to be fully general case. What is great is that all the hard stuff is implemented in this PR.

bhouston · 2016-05-19T17:16:26Z

/ping @MasterJames

bhouston · 2016-05-19T17:17:49Z

One aspect of polishing will be to get the light source radius implemented as previously described here: #7672

bhouston · 2016-05-20T00:47:49Z

Demo link: http://ci.threejs.org/api/pullrequests/8933/examples/webgl_shadowmap_pcss.html

bhouston · 2016-05-20T00:49:40Z

src/renderers/shaders/ShaderChunk/shadowmap_pars_fragment.glsl

+					numBlockers = numBlockers + 1.0;
+				}
+			}
+			avgBlockerDepth = blockerSum / numBlockers;


potential divide by zero? /ping @Spidersharma

MasterJames · 2016-05-20T03:48:45Z

Demo link cloth rendered black on my mobile but this is really great seems pretty quick too. Excellent amendments.
Maybe there's a need for an adjustible blur factor too.
Spot Light's penumbra?

bhouston · 2016-05-20T06:18:04Z

Did some more polishing and got it looking like this - smooth but grainy:

bhouston · 2016-05-20T06:19:22Z

@MasterJames wrote:

Maybe there's a need for an adjustible blur factor too. Spot Light's penumbra?

The physical solution is to have a radius of the light source. Larger bulbs make blurrying shadows, point-like light sources have crisper shadows. I'd prefer to go that route.

MasterJames · 2016-05-20T07:05:33Z

Sorry I thought the penumbra represented the same concept?

erichlof · 2016-05-20T14:49:23Z

@bhouston and @spidersharma03 ,
Thanks so much for this upgrade to the shadow system! One quick typo check, there is a double return statement on these lines (which I don't think is possible in glsl):
https://github.com/bhouston/three.js/blob/d1c22a07b826cb407032cfc05c69f944c8dfbc19/src/renderers/shaders/ShaderChunk/shadowmap_pars_fragment.glsl#L201-L202

Thanks again for the great contribution!

spidersharma03 · 2016-05-20T15:49:56Z

@erichlof Arh, yeah thats a mistake, and the second return should be removed. I forgot to remove that, though it should work.

mrdoob · 2016-05-20T17:02:16Z

Have you tried this on iPad? The fixed poisson disk usually makes things 2x faster on iPad (like 60fps instead of 30fps).

mrdoob · 2016-05-20T18:27:05Z

By the way, this looks really nice!

thmasn · 2016-06-30T19:21:00Z

will this eventually be pulled?
would work great with #9234

abelnation · 2016-06-30T19:39:19Z

FYI. The RectAreaLight work currently does not consider any shadow information. That's work to be done as a follow-up.

mrdoob · 2016-06-30T23:08:47Z

will this eventually be pulled?

Oh! I thought @bhouston said on twitter-land that a new PR was coming? Is this good to go @bhouston?

spidersharma03 · 2016-07-01T03:25:10Z

Yes, we did many changes in this PR, specially correct handling of directional light, and other bug fixes. I think @bhouston will be able to say better on the current state.

bhouston · 2016-07-04T14:21:47Z

I'd like to get this pushed, it is a pretty slow technique though. We ended up not using in in Clara.io for the time being because while it works and also looks great on high end machines, especially on GeForce 960+, it just kills low end devices.

mrdoob · 2016-07-12T01:49:12Z

I keep dreaming of how current FPS could be a value stored in the renderer and incorporated with policies that have pretty reliable defaults.

I've heard you. But I don't think the renderer should be that "smart". One could easily build a SmartWebGLRenderer on top that does what you propose.

MasterJames · 2016-07-12T09:00:45Z

Okay I wondered if you felt adding a super simple current FPS and average FPS (over a second or two) would be a useful idea in the main renderer or where. The one in stats seems excessive when needed? Thanks for your feedback and time as always.

mrdoob · 2016-07-12T23:15:28Z

I don't think the renderer should have any logic for that. The renderer/s should be as "dumb" and predictable as possible.

bhouston · 2016-07-13T00:52:45Z

I agree with @mrdoob, any sort of adaptive rendering quality should be
outside of WebGLRenderer, but I think it would be okay to put pieces of it
in examples/js. :)

Best regards,
Ben Houston (Cell: 613-762-4113, Skype: ben.exocortex, Twitter:
@exocortexcom)
https://Clara.io - Online 3D Modeling and Rendering

On Tue, Jul 12, 2016 at 7:16 PM, Mr.doob notifications@github.com wrote:

I don't think the renderer should have any logic for that. The renderer/s
should be as "dumb" and predictable as possible.

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
#8933 (comment), or mute
the thread
https://github.com/notifications/unsubscribe/AAj6_c7YdZ_2kHN70d6UJJ7PDQedc5t4ks5qVCBHgaJpZM4IiekW
.

deepak-ghimire · 2016-07-29T13:01:02Z

Hi Guys,
It's a great news that there has been some adaptation of PCSS algorithm. I had implemented a version , around a year ago, of PCSS for my mockup tool to generate final render (and preview) with PCSS shadows, but not too perfect :(

I am not a pro with GLSL but if some of my code portion is helpful please feel free to use.

It's a great news to hear that PCSS is progressing.
Many thanks to @spidersharma03, @bhouston , @mrdoob for the initiative.

you can find the code at Mockup Studio in
<script id="shadowmap_pars_fragment" type="x-shader/x-fragment">

P.S. apologies for a substandard comment, I am still new to GitHub.

spidersharma03 · 2016-07-29T13:11:03Z

I am preparing a PR for the PCSS, on which me and @bhouston worked some time back. Will submit it soon.

edankwan · 2016-09-29T10:13:39Z

@DEIP You version works very nicely. Can I hack it into a custom threejs build to use?

deepak-ghimire · 2016-09-29T10:26:53Z

@edankwan yes sure, Please note that I am using custom values rather than dynamic values, You'll have to consider that too.

edankwan · 2016-09-29T11:07:58Z

@DEIP I placed a directional light from the top, and it doesn't seem to work.

deepak-ghimire · 2016-09-29T11:11:09Z

@edankwan I had done it only for SpotLight,

ghost · 2016-09-29T11:15:23Z

We had handled directional light in a different and correct way by
introducing a spread angle parameter

On 29-Sep-2016 4:42 PM, "Deepak Ghimire" notifications@github.com wrote:

@edankwan https://github.com/edankwan I had done it only for SpotLight,

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
#8933 (comment), or mute
the thread
https://github.com/notifications/unsubscribe-auth/ABmpZozc3S27DuCRZuANiTmIr0cHq8ixks5qu50GgaJpZM4IiekW
.

edankwan · 2016-09-29T11:26:04Z

Sweet. Looking forward to the merge

sam-g-steel · 2017-01-21T19:42:46Z

@mrdoob, @bhouston,
I'm looking at updating this code and resubmitting this as a pull request that has no conflicts.

Would that be helpful?

mrdoob · 2017-05-25T23:37:58Z

@sam-g-steel yes please!

sam-g-steel · 2017-05-27T12:46:42Z

Cool, I think I want to finish the rest area light data ticket first... I think I found the perfect solution there! LZMA compression and Base64 I'll post more on that issue instead of here

…

On Thu, May 25, 2017, 6:38 PM Mr.doob ***@***.***> wrote: @sam-g-steel <https://github.com/sam-g-steel> yes please! — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#8933 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKdLJa44_vkVoOpjNpzyS13-y6_p5xcfks5r9hDygaJpZM4IiekW> .

mrdoob · 2017-08-16T01:37:41Z

Added it as an example... fc86899
http://rawgit.com/mrdoob/three.js/dev/examples/#webgl_shadowmap_pcss

mrdoob · 2017-08-16T01:37:46Z

Many thanks!

trusktr · 2017-12-13T02:42:42Z

Just for clarity, is PCFSoft different than PCSS?

ghost · 2017-12-13T03:00:19Z

Yeah, quite a bit. PCF simply applies a constant size filter to the shadow map, while PCSS filters based upon the distance between the receiver and occluder.

…

On 13-Dec-2017 08:13, "Joe Pea" ***@***.***> wrote: Just for clarity, is PCFSoft different than PCSS? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#8933 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABmpZtO85YeBZEwOQ1Dlvpw51FwKINnxks5s_znOgaJpZM4IiekW> .

MasterJames · 2017-12-13T09:15:01Z

For more lengthy information see...
PCF http://www.opengl-tutorial.org/intermediate-tutorials/tutorial-16-shadow-mapping/#pcf
PCSS http://developer.download.nvidia.com/shaderlibrary/docs/shadow_PCSS.pdf

trusktr · 2017-12-14T02:56:35Z

Oh, so PCFSoft lets us specify a constant shadow radius (blur) no matter of object distance, while PCSS is basically that but with increasing 'radius' as shadow goes further from the object?

ghost · 2017-12-14T03:00:12Z

Yeah, it tries to mimic the real world, where light sources always have a finite area, which produces penumbra and umbra shadow regions.

…

On 14-Dec-2017 08:27, "Joe Pea" ***@***.***> wrote: Oh, so PCFSoft lets us specify a constant shadow radius (blur) no matter of object distance, while PCSS is basically that but with increasing 'radius' as shadow goes further from the object? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#8933 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABmpZgz2txbbWpftLrxFtf2C4-mI2QMHks5tAI6pgaJpZM4IiekW> .

trusktr · 2017-12-14T03:19:56Z

Really why I'm asking, is because the Three.js WebGLRenderer Constants Shadow Types documentation says

PCFSoftShadowMap filters shadow maps using the Percentage-Closer Soft Shadows (PCSS) algorithm.

which seems to mean that using PCFSoftShadowMap will give us results like what is in this pull request, but that doesn't seem to be the case, so I was confused.

Is it just not officially implemented yet (other than in the examples), and PCFSoftShadowMap is a place holder for it?

mrdoob · 2017-12-14T07:40:22Z

@trusktr this is not part of the build at the moment, but it's used in a example:

https://threejs.org/examples/#webgl_shadowmap_pcss

bhouston and others added 3 commits May 19, 2016 13:07

PCSS First commit with shader. No poisson disk yet.

4b2edd2

PCSS Implementation

02283a7

polish PCSS example.

619476e

bhouston reviewed May 20, 2016
View reviewed changes

clean up the code.

ecbba55

bhouston force-pushed the pcss_remake branch from 5e5b2da to ecbba55 Compare May 20, 2016 01:41

bhouston added 3 commits May 20, 2016 01:52

replace fixed poisson disk with random one.

f5c4eec

more PCF samples than blocker.

5d33375

better sampling approach.

d1c22a0

Jozain deleted the pcss_remake branch November 28, 2016 15:21

mrdoob mentioned this pull request May 25, 2017

soft shadows that blur/fade with the distance? #11392

Closed

mrdoob added a commit that referenced this pull request Aug 16, 2017

Added Percent Closer Soft-Shadows example. See #8933

fc86899

mrdoob closed this Aug 16, 2017

ofrohn pushed a commit to ofrohn/three.js that referenced this pull request Jun 17, 2018

Added Percent Closer Soft-Shadows example. See mrdoob#8933

35ef617

mrdoob mentioned this pull request Feb 14, 2020

PCSS #9620

Closed

Percent Closer Soft Shadows by spidersharma03 #8933

Percent Closer Soft Shadows by spidersharma03 #8933

Conversation

bhouston commented May 19, 2016

bhouston commented May 19, 2016

bhouston commented May 19, 2016

bhouston commented May 19, 2016

bhouston commented May 20, 2016

bhouston May 20, 2016

Choose a reason for hiding this comment

MasterJames commented May 20, 2016 • edited Loading

bhouston commented May 20, 2016

bhouston commented May 20, 2016

MasterJames commented May 20, 2016

erichlof commented May 20, 2016

spidersharma03 commented May 20, 2016

mrdoob commented May 20, 2016

mrdoob commented May 20, 2016

thmasn commented Jun 30, 2016

abelnation commented Jun 30, 2016

mrdoob commented Jun 30, 2016 • edited Loading

spidersharma03 commented Jul 1, 2016

bhouston commented Jul 4, 2016

mrdoob commented Jul 12, 2016

MasterJames commented Jul 12, 2016

mrdoob commented Jul 12, 2016

bhouston commented Jul 13, 2016

deepak-ghimire commented Jul 29, 2016

spidersharma03 commented Jul 29, 2016

edankwan commented Sep 29, 2016

deepak-ghimire commented Sep 29, 2016

edankwan commented Sep 29, 2016

deepak-ghimire commented Sep 29, 2016

ghost commented Sep 29, 2016

edankwan commented Sep 29, 2016

sam-g-steel commented Jan 21, 2017 • edited Loading

mrdoob commented May 25, 2017

sam-g-steel commented May 27, 2017 via email

mrdoob commented Aug 16, 2017

mrdoob commented Aug 16, 2017

trusktr commented Dec 13, 2017

ghost commented Dec 13, 2017 via email

MasterJames commented Dec 13, 2017

trusktr commented Dec 14, 2017

ghost commented Dec 14, 2017 via email

trusktr commented Dec 14, 2017

mrdoob commented Dec 14, 2017

MasterJames commented May 20, 2016 •

edited

Loading

mrdoob commented Jun 30, 2016 •

edited

Loading

sam-g-steel commented Jan 21, 2017 •

edited

Loading