As an avid rock climber all the work in this video is incredible, mad respect to you and Jan for that adventure and then immortalizing it with the scan!
I don't do 3d work. I have no interest in doing 3d work. And yet I consistently become engrossed by these lengthy-ass tutorials just on the power Clint's specificity and obvious passion. True champ.
Having followed Clint from the beginnings of corridor back to this channel is such a mind-blowing experience. I just can't stop watching and being inspired by Clint in the same way he gets inspired by others. It feels indescribable
You get better results if you export two sets of images from lightroom. One contrasty for Model Generation, one delit for Texture Generation. It is possible to tell RC to use these sets :-)
From start to finish this was a fantastic tutorial, thanks Clint! Also being a rock climber this gets me excited to photoscan some of my favorite outdoor climbs.
Four minutes in, and as I expected, this video is very SPECIAL!!! I can't wait to re-watch it with better internet whenever I get home from vacations at the highest quality. It's clear how hard you worked on this video, and the result speaks for itself. Much love!
I WAS LITERALLY BORED AT WORK AND STARTED THINKING ABOUT PUTTING A LIDAR SCANNER ON A DRONE AND WONDERING IF THAT WAS A THING THEN I GOT HOME AND WAS RECCOMENDED THIS VIDEO. MY MIND IS BLOWN.
I needs to take out my drone, I'm hoping I can use it today to got a few aerials of the show I'm filming at (Inspecta Deck and Mastah Killa, Wu in the hooouse). Imma have to try this out. I blame you for my addiction to CG work hah. Peace out dude have a great weekend!
I'm currently writing on a thesis about photogrammetry in the construction industry and how to make target-actual comparisons. And AI is becoming a thing. Photogrammetry / LiDAR with drones is a huge new branch and the possibilities seem endless! I'm slightly hyped😎
Currently working on a AAA game right now and this is pretty close to how we’re doing rocks and cliffs. But also a lot of cleanup to remove vegetation and some zbrush work to make the asset usable in multiple angles
25:21 Note: normals should be unit length (i.e.: length of 1, normalized). Blending with a lerp will change their length, which can give unexpected lighting results. You could normalize the final values (so they're length 1 again), or you can use fancier math to rotate the normals rather than blending them. But really the main question is: are you sure you want to go through all the trouble of getting realistic normals only to flatten them again? Maybe you do, but you should probably at least think about it. Knowing the rules before breaking them and all that.
@@pwnisher I wanted to link to a relevant article, but RU-vid doesn't like that. Here's attempt 2 without the link. Best: leave them untouched, unlike bump maps they're already accurate. Medium: look at normal map blending techniques like the one described in "Blending in detail" by Colin Barré-Brisebois and Stephen Hill (a.k.a.: self_shadow) Better than nothing: chuck a normalize node at the end. Some normals may still be a weird angles, or 0, but at least the non-zero ones are at least the right length.
@@pwnisher Yep lerping a blue color isn't how to do normal strength, the blue channel on a normal contains no information it's essentially a 50% gray on the blue channel. Lerping in general should be viewed as a gradient switch to blend A and B, if the alpha switch is overshooting value between 0 and 1 you are essentially boosting the A alone in an unpredicting way. You should multiply by the same number your R and G channel then combine it back with it's blue channel, normalize the result and plug that in. Red and Green channel is what containing the normal lighting info in X and Y , which is why there are different formatting (directX OpenGL) based on the software axis system (y-up, z-up). To sum up on your example: (R channel * float "normalStrenght") (G channel * float "normalStrenght") then feed the result back in a "AppendMany" (where the blue come from the original one or just a number 0.5) to construct the resulting normal, normalize node before plugging in. You could also plug in a lerp between the boosted normal and the original giving you even more fine tuning controls as long as you clamp the control value between 0 and 1. Furthermore, desaturation node works better than choosing a channel from the albedo texture, since the real way of desaturation is to average the channels ((R +G +B)*.33). But this is almost accurate. Otherwise never ever use a 16k textures in a game engine, when unpacked in memory your 2 textures would be using around 2GB of GPU RAM. it would be really really bad in a gaming context you would have tremendous chugging in streaming and the amount of crash reports on your game would be unmanageable ;). People sometimes mess up their understanding of weight of textures in a realtime context, it's not the compressed size that matters if the file is 138mb it's not relevant at all apart from telling you the size it will require on disk, when opened in memory it's uncompressed and for textures it's ((pixelwidth * pixelheight * number of channel) *1.3 mipmapping) bytes.
@@2Jackrabbit Good addition about the uncompressed size! As for the R,G channels; I would expect it to be better to decode the normal, rather than trying to manipulate it in its compressed form.
@@RC-1290 No it would be totally fine since shaders works at runtime into what is uncompressed already ;) But that would still be the way to make a normal intensity, not a lerp with overshooting value in the alpha. That is something I teach about shader math, if you are "boosting" a lerp you either have the input wrong or using the wrong operation. A quicker way otherwise in unreal is to use the flatten normal node with a negative value.
Fantastic video Clint! Would love to see more stuff like this. It’s cool how this video was technically teased by Jan in the NERF finger video on Corridor. Also we’re only 7K subs away from 1M!!! Omg omg so excited for you Clint
Truly amazing work...my hats off in respect for your dedication and ingenuity. I never thought it could be possible to accomplish something like this. 👊😲👍
Looks pretty detailed! Very good scanning results. One issue I have with scanning big or medium rock-mountain geological formations are the overhanging parts - the bottom part of the overhanging rock is very dark because of the daylight shadow(not the sun shadow) and even with the new tool in Agisoft for delighting and extra Photoshop work- the texture is not really usable for high quality professional work.
well u would need to shoot at overcast days and also try to get a drone that can fit godox ar400 ringflash with a dslr so that there are no shadows there.
The fact that I found Clint's channel RANDOMLY just yesterday after being a Corridor sub for years is a crime. You gotta plug it on the Corridor channel, brother. At least once.
I think the ultimate would be to have a swarm of drones doing 3d scans. Like bees, but solar powered and linked to Starlink. Then you let them out into the wild and they scan the whole Earth bit by bit. Then AI is used to fuse and improve all those 3d scans together.
You can actually do a proper delighting pass in a completely different (be it somewhat complex) way that will give you 100% control of the new lighting of the scene. What you can do is take an HDRI from somewhere that would be mid air here I guess, in front of the mountain cliff. Then use that HDRI to light the model with a clay material (just plain gray without spec etc). Then you can bake that to a map (which would be your lighting map). Then go to your diffuse map, divide that by your lighting map to get your diffuse texture without lighting. It works really well, especially on things like these if you can get a proper HDRI (need some distance in this case probably, but most lighting is done by the sun and the sky, not much is coming back from the ground). Only thing you might need to add manually is a sun light if the HDRI doesn't capture it properly or you need sharper shadows, but you can get rid of 90-95% of the lighting in your diffuse texture EDIT: If I have time, I'll try to do this with this cliff and post the results in a week or so
Love the result! I can tell it didn't affect the final result much, which makes me wonder - how important is removing the harsh lighting in the original scan?
In my case, it wasn’t severe because the sunlight only hit like 10-15% of the scan (cause we got there so dang early) but I also chose lighting condition (in 3d) that favored the scan, as well as framed out the pre-baked lighting as best i could.
This is so cool! I've kinda wanted to do this with climbing for awhile too haha, I think someone is making an app using this too, like a 3d mountain project. I think it was called "rock garden" or something. So rad!
Good video! Nice to teach others! Side note Too many UV shells not optimal for real-time engines (break poly strips, limit optimal auto LOD, break GPU data stream). For good PBR material you need to take out shadows from the albedo - with AI.
Amazing tutorial, life changing for me (literally). I'm putting together my first drone setup and I was wondering 1) what drone model are you using and 2) what was the total flight time (for batteries)?
I need one more challenge! How about to make a looped animation, max 1 min long with ability to put it as well on horizontal, or vertical screen. Really hard to find a nice live wallpapers on android/windows
Hello friend, everything goes perfectly but the models are seen in black and white, when should I colorize or how can I have the coloring of the model done automatically?
I've been using my DSLR and running the photos through darkroom and meshlab, then editing in blender. Haven't done any scans with my drone yet, although I am planning to at some point. I'm definitely trying out this workflow when I do though!