Lemons Photogrammetry

[Bowcutt1]

Personal project to develop photogrammetry skills and refine workflow. High resolution models and textures were created using Agisoft Photoscan, then cleaned up using 3D Coat then retopologizing and UVW Unwrapped.

Rendered using 3DS Max & Corona.

Feedback/Critiques welcome.

Material Breakdown:

Thanks for viewing.

Thanks to www.bertrand-benoit.com for his guideance on workflow.

CREDITS:

Bowl: Design Connected 
Textures: Megascans / Arroway Textures
Process: Bertrand Benoit

[tomislavn]

Haha, wow! This is simply fantastic. Lovely renders and images are sooooo pleasant to look at! Beautiful materials and scene composition - really like it.

[clemens_at]

sweet! :)
very well done, although its a bit too much chromatic aberration for me. 

[Bowcutt1]

sweet! :)
very well done, although its a bit too much chromatic aberration for me.

Hi Clemens_at, Thanks for the comment.

RE: chromatic aberration, I Know, I should have dialled it back a bit really! Thanks for the feedback!

[cjwidd]

How did you model the lemon zest?

[pokoy]

The results are great. Maybe it's me but it looks like adding some simple sss would have taken the material way further in terms of realism. The geometry is great but for me the material lacks that last bit to be convincing.

[Bowcutt1]

How did you model the lemon zest?

Hi Cjwidd,

The zest was simply a box with a turbosmooth modifer, some noise, bend and other modifiers, optimised and then collapsed, there were 10 pieces in total, scatter using Forest Pack of Max object paint.

Hope that helps.

[Bowcutt1]

The results are great. Maybe it's me but it looks like adding some simple sss would have taken the material way further in terms of realism. The geometry is great but for me the material lacks that last bit to be convincing.

Hi Pokoy, Thanks for the feedback. There is SSS present, It's not perfect and TBH It needs more work. Next time I would use the skin shader and make use of the multiple layers. I found if I pushed it too far they looked like they were made of wax, which seems common in CG lemons. In reality there is a thin layer of very waxy skin (maybe 1.5-2mm) then you get to the pith which would have a different value for SSS and as it's bright white would (I think?) reflect allot of light back out. It's quite different to something like a strawberry where you have a very thin skin and then a largely similar inner flesh.

[cjwidd]

How did you model the lemon zest?

Hi Cjwidd,

The zest was simply a box with a turbosmooth modifer, some noise, bend and other modifiers, optimised and then collapsed, there were 10 pieces in total, scatter using Forest Pack of Max object paint.

Hope that helps.

Wow, very cool! Thank you for the explanation :)

[ASaarnak]

Very nice set of images.

[Juraj]

I would have clicked sooner if I knew there were these many images :- ). A lot more cool than I expected. Nice work ! Fun to watch :- )

When I think of CGI lemons, I always think of that SpecularNormals research paper which has shown how the proper surface can't be captured by neither photogrammetry nor pure diffuse albedo photometric stereo.
Only once they combined (which to my naked eye looked almost like 50:50) the albedo and specular normals, both from photometric stereo it looked correct. By then, the surface was heavily 'dented' which would be smoothed by SSS.

Watching your video, you in fact did all this :- ) In your own way (since I presume you overlayed generic noise, or highpassed noise from albedo scan?).
It's very close imho and I am not sure if it's the normals, or the SSS that betrays. I find Corona's SSS to be hard to tweak for weak/shallow effect.

I might try stronger displaced surface (from the denser noise that emulates specular normals) where the SSS could work stronger as well. Sorta like that snow shader from the Arnold tut around.

Anyway, don't mind my rambling, it's pleasurable work to see this :- )

[Bowcutt1]

Hi Juraj,

Thanks for the comments, really intersting! See my comments below in red

I would have clicked sooner if I knew there were these many images :- ). A lot more cool than I expected. Nice work ! Fun to watch :- )

When I think of CGI lemons, I always think of that SpecularNormals research paper which has shown how the proper surface can't be captured by neither photogrammetry nor pure diffuse albedo photometric stereo.
Only once they combined (which to my naked eye looked almost like 50:50) the albedo and specular normals, both from photometric stereo it looked correct. By then, the surface was heavily 'dented' which would be smoothed by SSS. Do you have a link to this information?

Watching your video, you in fact did all this :- ) In your own way (since I presume you overlayed generic noise, or highpassed noise from albedo scan?).
It's very close imho and I am not sure if it's the normals, or the SSS that betrays. I find Corona's SSS to be hard to tweak for weak/shallow effect.
The base mesh created by photoscan was quite noisy as it wasn't really able to resolve the pitting on the surface , not sure if a better camera would help here. I then extracted the pitting detail in photoshop, this was added to the normal map in 3D Coat. You are correct that I added generic noise in addition to the normal map as initially this was the missing detail. When you say 'High passed noise from alvedo scan' do you have any info on this process?

I might try stronger displaced surface (from the denser noise that emulates specular normals) where the SSS could work stronger as well. Sorta like that snow shader from the Arnold tut around.
I think I know what you mean do you have a link, I agree for the SSS to appear correct the mesh really needs to be displaced but I think the multi layer skin shader would help as well.

Anyway, don't mind my rambling, it's pleasurable work to see this :- ) Thanks for taking the time to read/comment!

[Juraj]

By highpassed detail from albedo scan I meant exactly what you now wrote you did :- ). It's the extracting of micro detail that scan didn't do. Highpassing because it removes gradients until you have only the detail you want.

Better camera wouldn't help because it's impossible to correctly scan materials that have strong SSS. The specular normal scanning was quite complicated and required motorized linear polarizer setup.
This shows a lot in Megascans too. They can't scan the fabrics properly at all.

But maybe spray-painting the lemons with matte grey paint ?

Here is image showing how strong the normals are actually when you separate SSS effect. It's lot stronger than what people think.

[Bowcutt1]

By highpassed detail from albedo scan I meant exactly what you now wrote you did :- ). It's the extracting of micro detail that scan didn't do. Highpassing because it removes gradients until you have only the detail you want. Ok cool, I will look into this more, as I think there is a better technique than the one I used!

Better camera wouldn't help because it's impossible to correctly scan materials that have strong SSS. The specular normal scanning was quite complicated and required motorized linear polarizer setup. Yep that does sound complicated, I am considering a polarised light setup but for now. 
This shows a lot in Megascans too. They can't scan the fabrics properly at all. Have you used Debarti capture much?

But maybe spray-painting the lemons with matte grey paint ? Yeah I have thought about this, would be interesting to see the kind of quality of mesh you could get from this workflow.

Here is image showing how strong the normals are actually when you separate SSS effect. It's lot stronger than what people think. Thats really interesting thanks, You've actually helped answer a few questions that I had in my mind about the limitations of the process, I partly thought It was my lack of skill but I see now there are limitations and I was correct to think I needed to add in addtional noise to help achieve the correct look.

[Juraj]

I've used only Substance photometric stereo node. It's the same as Dabarti just nicely integrated.