Author Topic: Technical question - 3d scans by Poliigon  (Read 9629 times)

2018-11-27, 22:24:14

Hadi

  • Active Users
  • **
  • Posts: 82
    • View Profile
    • Behance
Hey guys,
I've been testing some new 3d scans with the guys at Poliigon these days and we can't replicate the quality of the details visible with Blender and Cycle in Corona.
Basically, main issue is the fact that the diffuse map is losing those small little details visible in their demo render, other than looking a lot more rubbery/glossy than the way more realistic one done in Cycle.
I've attached some screenshots for you to have a look at. Let me know if you need more info.

2018-11-28, 01:40:14
Reply #1

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 6132
    • View Profile
Any chance of sharing the material set ups? A look at the renders doesn't look to me like a difference in the Diffuse channel, but in the Glossiness, Fresnel, or some other reflectivity attribute. Could simply be a case of the material not being converted properly, rather than any inherent loss of detail in a map.

EDIT - I am not sure what the third image shows, so can't comment on that :) Just looking at the "looks more rubbery" and the clear differences in reflectivity between the two renders.
Tom Grimes | chaos-corona.com
Product Manager | contact us

2018-11-28, 07:22:08
Reply #2

Hadi

  • Active Users
  • **
  • Posts: 82
    • View Profile
    • Behance
Hey Tom,
The third image is to show the small details in the diffuse map that are disappearing at render time.
I will speak with the guys at poliigon to ask them if it's possible to share the model with you.

2018-11-28, 10:37:02
Reply #3

Hadi

  • Active Users
  • **
  • Posts: 82
    • View Profile
    • Behance
Tom, I had a reply from the Poliigon team who did some further test dropping the IOR value. You can see the result in the attached screenshots.
i think thought than an 1.1 IOR is incorrect as, according to the PBR workflow, the reflections should be fully white, without a map in the reflection slot and a 1.52 IOR.
Let me know your thoughts about it.
Thanks.

2018-11-28, 11:14:47
Reply #4

sprayer

  • Active Users
  • **
  • Posts: 803
    • View Profile
if you in 3ds max try to reduce blur filtering in bitmap loader

2018-11-28, 19:13:29
Reply #5

Hadi

  • Active Users
  • **
  • Posts: 82
    • View Profile
    • Behance
Ok, I think I've nailed it down and I'm reporting here in case someone is interested.
You can check the screenshot, but I think the result is quite close to the demo render from Poliigon, and I've just used a studio setup.
So, IOR 1.52, PBR mode on but with the reflection map in the slot, toned down by 50% of the one Poliigon provides with the model. I know that's not the correct approach, but I'm having a way better result with the map on. I'm open to suggestion if I'm doing something wrong.
Normal map bumped to 2.5 with the flip green ticked on.
Reducing blur filtering didn't change anything.
Reflection and glossiness maps have been loaded with the automatic mode, but I would like to have your opinion if this is wrong and I should load them with a 1.0 override.

2018-11-29, 13:04:41
Reply #6

Jpjapers

  • Active Users
  • **
  • Posts: 1716
    • View Profile
Ok, I think I've nailed it down and I'm reporting here in case someone is interested.
You can check the screenshot, but I think the result is quite close to the demo render from Poliigon, and I've just used a studio setup.
So, IOR 1.52, PBR mode on but with the reflection map in the slot, toned down by 50% of the one Poliigon provides with the model. I know that's not the correct approach, but I'm having a way better result with the map on. I'm open to suggestion if I'm doing something wrong.
Normal map bumped to 2.5 with the flip green ticked on.
Reducing blur filtering didn't change anything.
Reflection and glossiness maps have been loaded with the automatic mode, but I would like to have your opinion if this is wrong and I should load them with a 1.0 override.

Can you show a material graph for your approach?

2018-11-29, 13:16:09
Reply #7

Hadi

  • Active Users
  • **
  • Posts: 82
    • View Profile
    • Behance
Ok, I think I've nailed it down and I'm reporting here in case someone is interested.
You can check the screenshot, but I think the result is quite close to the demo render from Poliigon, and I've just used a studio setup.
So, IOR 1.52, PBR mode on but with the reflection map in the slot, toned down by 50% of the one Poliigon provides with the model. I know that's not the correct approach, but I'm having a way better result with the map on. I'm open to suggestion if I'm doing something wrong.
Normal map bumped to 2.5 with the flip green ticked on.
Reducing blur filtering didn't change anything.
Reflection and glossiness maps have been loaded with the automatic mode, but I would like to have your opinion if this is wrong and I should load them with a 1.0 override.

Can you show a material graph for your approach?

There you go.

2018-11-29, 13:20:53
Reply #8

Jpjapers

  • Active Users
  • **
  • Posts: 1716
    • View Profile
Nothing seems off apart from the bump being at 2.5.
I wonder if this is anything to do with the way corona samples bump maps and doesnt do microgeometry well.

2018-11-29, 14:15:58
Reply #9

romullus

  • Global Moderator
  • Active Users
  • ****
  • Posts: 9283
  • Let's move this topic, shall we?
    • View Profile
    • My Models
Nothing seems off apart from the bump being at 2.5.
I wonder if this is anything to do with the way corona samples bump maps and doesnt do microgeometry well.

It does not. Normal maps should almost always be set to 1.0 strength and Corona has no problem with that.
I'm not Corona Team member. Everything i say, is my personal opinion only.
My Models | My Videos | My Pictures

2018-11-29, 14:16:46
Reply #10

Jpjapers

  • Active Users
  • **
  • Posts: 1716
    • View Profile
I wonder if this is anything to do with the way corona samples bump maps and doesnt do microgeometry well.

It does not. Normal maps should almost always be set to 1.0 strength and Corona has no problem with that.

My bad i just noticed its a normal not a bump

2018-11-29, 15:13:48
Reply #11

Hadi

  • Active Users
  • **
  • Posts: 82
    • View Profile
    • Behance
Nothing seems off apart from the bump being at 2.5.
I wonder if this is anything to do with the way corona samples bump maps and doesnt do microgeometry well.

It does not. Normal maps should almost always be set to 1.0 strength and Corona has no problem with that.

At the end of the day what matters is how it looks. So if breaking the rule makes it look better, I would use a 10 multiplier.
I actually had to do that in order to make the bump pop out from another model of theirs (the blueberries), so I'm not really against on breaking the rules when necessary.

2018-11-29, 17:02:48
Reply #12

romullus

  • Global Moderator
  • Active Users
  • ****
  • Posts: 9283
  • Let's move this topic, shall we?
    • View Profile
    • My Models
If you have to greatly exaggerate normals just for model to look right, that might mean there's still something wrong with your setup, or with textures.
I'm not Corona Team member. Everything i say, is my personal opinion only.
My Models | My Videos | My Pictures

2018-11-29, 17:07:28
Reply #13

Hadi

  • Active Users
  • **
  • Posts: 82
    • View Profile
    • Behance
If you have to greatly exaggerate normals just for model to look right, that might mean there's still something wrong with your setup, or with textures.

There's definitely something wrong with the textures that came out from the 3d scan, as I had to manually correct all of them (refl, gloss and normals).
As I said before, I had to boost 10 times the normal map in the blueberries in order to get the details out of the normal map.
I am not sure if the original one rendered in Cycle was actually looking correct without any map being corrected.

2018-11-29, 23:37:46
Reply #14

romullus

  • Global Moderator
  • Active Users
  • ****
  • Posts: 9283
  • Let's move this topic, shall we?
    • View Profile
    • My Models
They don't have scanned models free sample, so i downloaded this texture, which is also scanned, so should be good indicator on their texture quality. To be honest, i found nothing wrong with those textures - once you plug them in material, it works as expected. No need to boost anything, well... maybe except glossiness. That map is good by itself, but it's for dry mud and in their preview, mud looks wet, so if you want to replicate that, you need to make that map lighter. CoronaColorCorrect node is perfect for that. One thing is worth to mention, they also provide reflection map, don't know what's its purpose - i didn't use it. Instead i plugged into reflection slot AO with boosted contrast - nice trick advocated by dubcat and Juraj Talcik.

I'm attaching test renders.
01 - maps are pluged without any modification (AO and reflection aren't used)
02 - glossiness map is plugged through CoronaColorCorrect and brightness with gamma are increased
03 - same as 02, but high contrast AO is plugged into reflection slot
I'm not Corona Team member. Everything i say, is my personal opinion only.
My Models | My Videos | My Pictures