WebGL Path Tracing

[michalis]

Abn, blender isn't half finished. Cycles is. There is the internal render and this is what we use for baking maps.

Cycles is a modern engine, a pathtracer. Normal maps support is almost ready but I doubt if it will work as expected.

Pathtracers actually suffer from a serious issue. It's called terminator issue. Smoothed shading doesn't look as expected. Under hard lighting, low density geometry becomes visible in areas between light and shadow. This ugly effect makes practically impossible to render a low poly figure with some normal mapping.

On the other hand pathtracers render fast, so a hi density poly can be used for baking maps. It's a great solution as AO and GI are superior to scanline based renderers.

We have to wait and stay tuned with this new generation of render engines.

Abn, you love Nvidias, cycles will make you feel happy. It's very fast under CUDA.

There's octane too. his excellent engine also suffers from the above limitations.

Abn, you should not forget. Render engines aren't for video games only. You probably know better than me that cinema industry never uses normal maps.

I think that the basic reason is that normal maps don't react with GI and AO well, or should I say, at all.

[AbnRanger]

Normal maps have basically replaced grey scale bump maps, no matter what medium is used. Most will use displacement and Normal Maps (for the uber high detail).

[michalis]

You are wrong.

This "most" indicates it. Sorry Abn but you can't go for cinema production with normal maps.

There're lot of issues against normal maps. You probably know it better than me.

The future game engines may not use them.

[TimmyZDesign]

Michalis I agree with you. I think the future of rendering is definitely in realtime engines with pathtracing. Blender's Cycles is great, as well as others such as Octane. My graphics card only has 192 CUDA cores, but the renders it produces are still very fast. I am probably going to buy an additional GTX 560 to take even more advantage of this new technology. Marmoset Toolbag is also great! It supports normal maps and it is realtime as well! It is really a wonderful feeling to be able to change settings and immediately see the results...instead of constantly waiting for the render to finish! I also found some tutorials that suggest that it is actually possible to use normal maps with Cycles...have you seen them? I have not had time to try it yet...

http://www.youtube.com/watch?v=eFO4TIezS-o

[michalis]

I also found some tutorials that suggest that it is actually possible to use normal maps with Cycles...have you seen them?

Of course I have seen them.

They use the blue channel as bumps. It's ridiculous. Please, don't spent your time with such nonsenses.

Normal-maps support implementation is almost ready. So, just wait. How good it will be, I can't say. I have my reasons to doubt on this.

There're more goodies to come. Like micro displacement. There's much more potential on such approaches.

BTW the above video tutorial is fatally wrong. You must avoid to use color mix on a value that's a B&W image. I doubt if even supports 32 bit maps.

So, for the displacement node you need a mathematical node, as multiplier. It is called displacement but for the moment it's just bumps.

There's an experimental mode, under render panel. This enable real displacements but it's very buggy and unfinished.

Your best workaround is to import a displacement map and apply it into displacement modifier. In conjunction with a subsurf modifier. Nothing to do with cycles, it provides excellent real displacement.

Now, if you subdivide (let's say) x3, go back to 3dcoat or other app (even blender) and try to bake and use a second displ map, baked from this subd level. Use this as bump map for capturing Hi freq details.

A valid method, the best IMO, very well presented in zbrush forum. A workaround for quality renders, under all known apps.

Another one blender sculpting, just for fun. The other two figures for my whole composition. What are they doing? They take part on something... what? If I manage to finish it, if it comes as I expect, you wont ask. (lol, I'm not that good, just kidding)

[Tony Nemo]

"What are they doing? They take part on something... what?" I think they are watching a game of chess.

[michalis]

A game of chess !

Not really.

But I love chess and I don't have many opportunities to play these days.

Only against these silly machines. Finding lot of tricks to win them. There isn't much fun on this.

This isn't the case. This is another game taking place. Not a good one.

[pixo]

For the normal map issue there is another really popular path tracer that don't support it (even if they are fixing it), no need to blame Cycle.

[michalis]

I don't blame cycles, for not supporting normal maps yet.

IMO there's not much future in normal maps. The issues are too many and modern render engines (pathtracers) don't like them.

One of the biggest issues is that normal maps don't react well under AO and GI. Displacements do work well, also react well when used as bumps.

So, when using normal maps we also bake AO maps. This is the main reason. This is also what 3dcoat does in first place.

Cycles will support normaps but it will also be capable to bake renders on UV maps. It also has to find a solution on terminator issue. This is the most difficult. Most pathtracers suffer from this "disease"

[chingchong]

i dont think that pixo meant you

[pixo]

I confirm chingchong, i was just trying to say that cycle was fairly good.

By the way the terminator issue is not a big problem, subdivision correct it,by the way when you model properly your surface you don't need to subd a lot.

Kind

[michalis]

Subdivisions makes it less visible. It's there however.

[pixo]

Nope it doesn't happen that much in most path tracer.

At least not the ones that i tested 3 without cycle.

It happens only if you are using oldschool lights.

The one that emit infinite energy from one point, aka spot,omni,etc light without any size.

With most probably a very old shadow function.

If you need these kind of lights ,you should probably use a Reyes with micropolygon subd it will be better.

Especially because you have way to offset the ray(bias) in an intelligent way .

Kind

[michalis]

It happens only if you are using oldschool lights.

Old school lights? lol

You probably mean hard lights. Like sun in an exterior for instance. It's a problem for architectural visualization. It kills smoothed bevels under a certain angle of sunlight.

Indeed, using large lights-smooth shadow casting helps a lot. But you need hard lighting in half of the cases.

[pixo]

For Sun light you can use IBL with MIS you can get easily sharp shadows without this problem.I did it many times.

Of course you will get more realistic result too.

[pixo]

By the way if you really want to use old school lights , just bias you ray and it works.

[pixo]

just to make you happy.

I did path tracing with a REYES and Raytrace, directionnal light with a shadow at 0.5 degree raytraced shadow of course.

The sphere is really low poly, i just have to bias a little bit the ray and there's is no problem.Old school method for old school problem.

Take care.

[michalis]

Thanks pixo.

But, cycles is not a biased renderer. Neither octane.

Another older unbiased renderer is Kerkythea. And thea. They found a biased solution on this problem though.

[pixo]

Bias mean offset the ray in this case, i m not talking about convergence of the integral.

I used a biased Reyes ,and an unbiased Raytracer to show you that's working in both situation.

By the way you can do path tracing in both ,Raytracer and Reyes,and use unbiased render method in both type of renderers.

You probably misunderstood the unbiased meaning .

[michalis]

How low poly the sphere is, I can see (in the shadow casting)

I wonder, is it so simple? I keep complaining on this, in blenderartists forum, among other users. Cycles can't do it. For the moment. Neither octane can.

Are these methods support GPU-cuda or maybe OCL? Are you sure? This GPU support became a bottle neck on development.

[pixo]

Yes Michalis it's that simple hopefully for us 2 .

You just need to add a raytrace bias to your objects if cycle consider per object attributes or to your shadow light function which mean to your light attribute .

If you don't find it , just complain to don't have shadow raytrace bias or raytracing bias.

Hope it can help you.

Take care

[pixo]

Are these methods support GPU-cuda or maybe OCL? Are you sure? This GPU support became a bottle neck on development.

For sure since it's nothing else than displacing the position of the ray origin along the normals.

I insist your result will not be biased at all in term of render.

[michalis]

Now, I have to post such ideas on blenderartists forum. They will eat my alive but it won't be the first time LOL

[pixo]

If cycle doesn't have it , you are right to ask them since every render have this at least all that i tested.

It doesn't solve everything but it helps a lot.

[michalis]

Basically it's a dirty trick. Not for purists, if you know what I mean LOL.

But, who cares...

[pixo]

It's isn't a dirty trick...or subdivide your model to the micro-polygon level.

Actually if you are such a path tracer's fanatic you would never use these kind of lights .

Even solid angle consider this kind of what you call "trick", and they are the hardest fanatic that i saw,they didn't even consider "bake" or "PTC" as a solution .

Using a light with infinite energy that your are multiplying by a shadow function with a different distribution than the surface brdf ,it is not what i call path tracing purism.

So there's no purist in production at least ..... IMHO if you wants to go further ,path tracing should be very used for exterior shots with IBL and MIS this is the only case where it's really efficient (my 2 cents biased renderers does it very well for less computation time).

But for serious things like interior ,you should use BPT cause it's predictable without any bias and far more faster than PT.

MLT is elegant, converge faster but doesn't allow a fast preview since you have a start-up "bias",the problem is that your lighters will wait more before to have a proper scene lighting preview , for instance Maxwell renderer use on a thumbnail , for me it is not enough.

My 2 cents, just keep in mind that we hopefully have tools that help us .We should use them properly.

I can give so many cases where path tracing can't do something at reasonable cost ,this is the same for the others methods.

There's not today an absolute method that's it .

Keep in mind that PT with MIS is very old at least older than the modern software.

Take care

Pixo

[michalis]

Dirty is not what I think. It's what cycles developers think about it.

I like dirt.

[TimmyZDesign]

Very interesting thread. Thank you for the information about rendering!