Ambient Occlusion Shadows Removal by GrzegorzBaran
Вставка
- Опубліковано 25 тра 2019
- #photogrammetry #scanning #shadowremoval #pbr #workflow #delighting
The video I made to show the way I remove ambient occlusion shadows from scanned surface using Substance Designer.
As an example I picked the cliff surface I scanned recently.
In details I covered:
- how I used to remove shadows using 'AO Cancellation' node in Substance Designer
- how I do it now using baked Ambient Occlusion texture
This type of shadow removal might not work for every type of surface but for surfaces like this one works great.
For those interested getting even more into photogrammetry details, here is a link to my photogrammetry guide:
gum.co/YanD
Thanks to those who bought it already, as it really helps me to move forward with my research and create even more content.
This video is also available as offline version on my gumroad:
gum.co/ahIFI
By buying it you support my research and help me to create and share even more content like this one.
Feel free to follow me on Artstation:
www.artstation.com/gbaran
If anyone feels that captions would be helpful, please leave the comment and I will add them.
Hope you have found this video helpful,
Thanks for watching,
Grzegorz Baran
Another excellent video, very clear and concise, thanks for sharing i look forward to the next one
Thanks Barny :)
Awesome ! Can be very usefull
Cheers :)
Thanks for sharing, very useful technique.
Thanks man, I am super happy to hear that :D
Really smart. Does the color equalizer tile correctly? Last time I tried it I thought I saw artifacting at the seams.
THanks, I never had any tiling issues with color equaliser.. anyway, I dont use this technique anymore for scans as I developed much better and accurate solution and work directly on raw data - but this one is still an option for some hardcore cases :)
@@GrzegorzBaranArt thanks for the reply. It's reassuring. Maybe all it did was reveal the tiling errors I already had! or maybe because my version of substance (2020) is old. I'll give it another try.
It turns out, the Color Equalizer node is unique, it requires you to check a box instead of using the tiling settings. How infuriating. It's especially annoying to me because I like to scan trees, for tree bark material - and since they already tile on one side (since the tree is a cylinder), I don't want color equalizer to tile vertically! But the check-box doesn't have the option to tile horizontally, vertically, or use both, the way _everything else_ in Substance Designer does. 😵
@@josephbrandenburg4373Yeah, you need to set the 'Tiled Input' flag if the input is already tiled. As said, I dont use this node too often as I found much better and less data destructive ways to handle such cases. Equalising isnt free and affects quality. Its a math based one so its pretty dumb and while can do the job for subtle cases, I found this quality loss totally unacceptable for more extreme cases - hence finding ways to reduce need of such operation - increased capture consistency, color calibration and direct pre-editing of raw-data before reconstruction etc.
I was using this technique with Photoshop and it helps. Do you think using photogrammetry and then the AI to Material(Alchemist) just for the albedo works better?
Yes, using Alchemsit delighter just for the albedo makes sense same as generating roughness from this Albedo can save some time and should improve the workflow.
@@GrzegorzBaranArt Ty Gregorz I thinked about that watching your vid, and I said I should ask Gregorz :)
Good job, I seen your new method on Arstation. I don't use AO cancelation node as he make textures too bright. I will try your method for compare them. I'm personnaly using an Object Space Normal map method, do you know this technic ? It's used by several studios and do a very good job.
Keep creative.
Thanks mate. Looking forward to hear how it works for you :). Nah, I didnt try Object Space Normal map, thanks for pointing this out, will give it a try :). Cheers!
btw. the big income of my technique was that it didnt just increased the brightness of dark areas as it would not work very well due to the lack of data. This technique helped me pretty easily to fake this data and it is something which brings value :). Usually its something what AI should really do and I do hope it is going to one day.
Maybe you can get the information of dark area by higher your exposure? You can try use combined HDR image to generate material (I haven't tried it yet. I'm just guessing).
Yeah, this is a good idea and totally correct way of thinking. I think have found quite good and efficient way to remove ambient occlusion shadows.. its quite simple.. just testing it before I make it public to make sure it works well without any issues :). And yeah.. HDR would be perfect.. but it means longer exposure and in result more blur. But for sure this is the way to go with really .tough surfaces which cannot be covered by available dynamic range.
@@GrzegorzBaranArt Thank you for your answer :) , I look forward to your research results!
How did you get a roughness map from a scan?
Albedo looks great
Thanks, I don't. The roughness in here was generated in Artomatix.. should mention this in my tutorial probably but thought it is irrelevant here and if I would mention too many things I would lose focus. Wanted to keep it short .. my previous one was over 1h long :). So If you watch my previous one you should get more insight into how I make a roughness maps:
ua-cam.com/video/3bGwGpLpSa4/v-deo.html
Hope that helps, Cheers!
@@BoogerBrain You can't generate this map with "standard pbr photogrammetry" approach. You have to capture additional photos with special lighting like polarized light for example. Also you can fake the roughness from the diffuse and ao maps in substance. Not sure that Artomatix make a great job without special ligthing for generate the Roughness map in all conditions.
that s not how you should do it. It takes time, it s not pipe friendly and there s a better technique that takes like 3 seconds and gives way better results with no dark spots
Sounds awesome Yann but kind of negative, could you please share your technique then? It would be very appreciated by everyone for sure.
In this video I said that this is one of many ways for albedo delighting and the one presented is just as an alternative to 'AO Cancellation' node. I shared it since I have found that I am getting quite good results compared to my previous approach and thought someone might find it useful :)
@@GrzegorzBaranArt yeah not using substance to cancel it. Just take ur albedo to photoshop, ur occlusion map, make it a mask, use content aware remove tool and play with alpha edges. Done
@@yannmassard3970 oh. thanks for sharing :). I would say Photoshop is an awesome tool but is a kind of a step backward nowadays. It works with single maps but doesn't on more complex materials. I didnt want to complicate this graph but what if you add some patching and want to have it processed on all channels with live material preview?
Ambient occlusion delighting is just the one of a few tweaks I usually do in Substance. When you create graph you can easily turn it into a single node and reuse it in a future with different materials. If you keep the naming convention you don't even have to re-apply any connections as it can be auto-read. It means that if you do it 'my way' you can auto-delight albedo in just a few seconds - time needed to open the app and press the export button .. eventually do some minor tweaking. You just need to create initial graph and simply save it to be reused.
I have a few more ideas for delighting like delighting on early stage using wider tonal range coverage ..before even initial reconstruction happens and a few more but want to research them a bit deeper before I record any new video to share my results.
Hope it makes sense even if you don't agree with me and big thanks for creative response :). Have a nice day
@@GrzegorzBaranArt
Hi Grzegorz,
I am following your inspiring tutorials all the time. You are an extremely creative and experimental artist. The AO removal you are showing here is another feature which I just learned from You. Keep it up!
@@maciejjanikowski3622 Thanks a lot Maciej! It is I super cool to hear something like that :). Cheers!