It's quite a big tutorial, the timeline may come in handy: 00:00:00 - Start 00:00:01 - Introduction 00:01:15 - Smartphone photos 00:01:47 - Pre-precessing the RAW photos in Darktable 00:09:13 - Reconstructing the 3D scan in Reality Capture 00:14:11 - Simplifying the mesh 00:15:18 - Texturing 00:18:26 - Clean-up the scan in Blender 00:22:36 - Boolean operations 00:33:04 - Wheels 00:35:15 - Windows 00:36:28 - Knife project 00:38:36 - Drawing topology 00:49:47 - Headlights 00:51:02 - The Knife tool 00:53:20 - The truck roof 00:55:47 - Denoising the highpoly mesh via Blender sculpting 01:02:21 - Export to Reality Capture 01:05:38 - Bake textures in Reality Capture 01:09:07 - Creating materials in Blender 01:14:12 - Fixing the roof (clone brush) 01:17:52 - Upscaling the diffuse texture using neural network (CupScale free tool) 01:21:39 - The last minute tweaks 01:25:13 - Conclusion
Красавец, четкий туториал,всегда двигал точки (vertexes) рядом с окнами машин в ручную чтобы потом применить материал и даже не догадывался что можно просто сделать кнайф проджект🤙🤙🤙
I was always curious about photoscanning, but after seeing all this process, I think it's still easier and less resource-consuming to just make your own model from scratch.
Depends on the subject matter, for things like tree stumps and organic shapes, photoscanning really shines. For hard surface stuff it's not the best and needs a lot of cleanup
I would say photogrammetry is still good for hard surface stuff when we are talking about something as destroyed or even more destroyed than this truck. It's very difficult to build a perfect hard surface model and make it look worn out or destroyed and look natural at the same time. Or maybe I just don't have the patience for that. Imagine a rusty destroyed tractor still filled with hard surfaces. But will look much better via photogrammetry than trying to achieve something only by referencing photos and texture painting.
Great tutorial and the end result is awesome. But that's a loooot of work and would probably take me three times as long as actually building such an object from scratch and texturing it with the textures instead. Not saying that my end result would be as good though..
The model in the end doesn’t look bad. Definitely usable for something in a background. I think the one thing that bugs me about it is the glass. The waviness of it gives off a red flag that something isn’t realistic as the glass would either be in one piece and smooth, or broken and missing or with holes. I think it’s best to just remove the normal map from those objects since they would have a very uniform smoothness, and use the roughness to give some imperfections. Throw on another image and use a color ramp to define which parts have the dirt and grime.
Anthony, I agree, generally we would expect the glass to be more or less undisturbed by normal map. That being said, for some reason which I can't put my finger on it looked more believable *with* a normal map (not even with a normal map, but with an EXTRA large-scale noise applied to object normals - as seen in the project files). At least, that's my perception of it. In one of the test I even exported a broken mesh from RC (it has 'fixed' geometry by making it manifold, which turned the one-sided objects into a self-intersected mess). And this broken self-intersected geometry looked fairly believable when assigned the glass shader, maybe due to fake depth effect introduced by mangled normals. There's something to it, I just don't know how to describe it properly I guess :)
We felt that the 'smartphone capture and then fixing the mess' workflow didn't get the attention it deserved, so this is our attempt to fix it somewhat. :)
24:06 what I like to do is select what I want to keep with select linked (hover over the part and press L), then invert the selection (ctrl i), and then delete. Edit: looks like you do end up mentioning this method at 30:15 for getting a single vertex from a plane. For this, there is a built-in addon that adds a single vertex to the add mesh menu. I can't remember which addon it is though.
that knife project shit was the thing I was waiting for for 36 minutes and for the life of me I can't do it. I repeat your steps and it doesn't project or projects wherever it wants 😄 I did it the right way by chance once and can't reproduce it now for 15 mins. Some things are just not to be. Reminds me of my high school math classes. Thx for the vid though. I'll still learn whatever I can from it 🥲
Thanks for the wonderful lesson! Can you please tell me how to make an ordinary polygonal model with the correct mesh from photogrammetry of a car model? Perhaps there are some special programs for this? I would love to watch a video on this topic.
Such an awesome tutorial! I was wondering... For the interior... Is it just as easy to get rid of all the garbage inside and whip up some basic shapes that somewhat resemble car seats, steering wheel etc instead of leaving it as it is? Or.. Is that a whole ton of other work? You make everything look so easy and Im thinking Ill hafta watch this about 10x (or more) to "get it" .
What would be a good method to create very fast trash models of the cars. The faster, the better. I would use those for our car wrappings, which I sometimes design in my job for our clients. Lidar?
@@GlebAlexandrov yeah, I think that makes sense. My biggest worry is to save the money of the clients so that we don't need to bill them too much of work that can be done somehow faster. But this 3d-thing quite much is just an addition. The whole process obviously can be done without using any 3d softwares. It just is cool that we can (if needed or wanted) show the end result in 3d before physically touching the car. 🤔
Hey. Just downloaded the video onto my PC. When you make a 3D truck, how do you work on putting in the interior such as the dash, shift lever, and a control on how to activate the windshield wipers, etc.?
Honestly - I think it would be easier and better to just do the retopo of this model :D Unless you REAAAALYYYYY don't want to mess around with texture projection and so on :D
I have followed the tutorial step by step. At timestamp 1:06:13, reimport back into Reality Capture I get an error, "The model contains non-textured triangles." What have I done incorrectly? (Great video by the way, I appreciate your providing it as I have been wanting to watch your Creative Shrimp photogrammetry course- just not in the budget at this time).
Есть ли способ и смысл показать программе какие точки на фото стоит учитывать и почистить ошибочно определённые совпадения? Например, если фигура является созданной в параметрическом редакторе машиностроительной деталью и состоит из ровных плоскостей, окружностей, углов предсказуемой величины, тел вращения и тому подобного, может просто отметить на фотографиях несколько точек вручную по углам и вдоль кривых?
Could it be reconstructed with pictures that are downloaded randomly from google image search? .. (then some duplicated and mirrored to represent the other side of the vehicle)
Do you mean by the means of photogrammetry? I don't think so. Otherwise, using the random pictures as references for the regular modeling so to speak, yeah possibly :)
Yes and no. I think it would be slower to do a retopo, but final result would look much better. But it's not necessary for a BG model that will be somewhere beyound level bounds or just out of player's/viewer's reach. That's not a way to do a main focus object tho. For that you'd have to ideally get a blueprint of a car model it with all the details you can see. Or at least do a proper retopology + model in a few details like shock absorbers axles etc.
wiem że jesteś polakiem bo tablice z elbląga na busie to napisze po polsku, mordo link przestał działać do pobrania plików projektu jeśli by sie dało to jakoś odnowić czy coś byłbym bardzo wdzięczny bo chciałem ozbaczyć czy mój procesor sobie wgl poradzi z odpaleniem takiego projektu w blenderze
hmmm ... looks like you can model this truck from scratch by using reference images for such amount of time.. what you may need that bad mesh for? a cinema VFX prop? Looks completely useless for gamedev (for hard surfaces tho). Rocks and other natural objects i think will be good enough.. but cars? idk...
A background asset most likely, I'd say. At least that's my perception of it. Out of curiosity, how much time you think it would take to model something similar from scratch, based on reference photos? I assume you mean the photos from internet, or do you mean front-side-etc projections?
yeah, I mean Front\Side\Rear view projections and not very detailed result where you can simply use your photos as basic textures. You will end up with smth close to the result of "3d scanning" (or however it' called)) but more versatile in terms of use. For example that wheels I suppose are easier and faster to be modelled from cylinders than cleaned up on a 3d scan)
@@GlebAlexandrov you could block that out in moments and add detail in a second past far faster than your method. Although very interesting nonetheless
what phone takes raw format photos? with that you lost me already...not sure it wouldnt be faster and cleaner to just 3d model something and sub painter it and be done with it. this is not really clean enough for a game. and takes almost the same amt of work or more than a more traditional approach. seems like this process is not quite there yet.
I feel bad for losing you. Some phones support RAW out of the box, some need an app to shoot in raw (correct me if I'm wrong though). Open Camera app for Android for example (not all cameras support it still).
This is finally a tutorial that covers almost everything. Skipping over details is what would make me not understand certain parts. You can always speed up if you want and skip certain parts
I know what you mean, it's that kind of vibe for sure. It's Poland though, but let's say, it's in the same Half-life 2-esque universe when it comes to abandoned trucks :)
It's quite a big tutorial, the timeline may come in handy:
00:00:00 - Start
00:00:01 - Introduction
00:01:15 - Smartphone photos
00:01:47 - Pre-precessing the RAW photos in Darktable
00:09:13 - Reconstructing the 3D scan in Reality Capture
00:14:11 - Simplifying the mesh
00:15:18 - Texturing
00:18:26 - Clean-up the scan in Blender
00:22:36 - Boolean operations
00:33:04 - Wheels
00:35:15 - Windows
00:36:28 - Knife project
00:38:36 - Drawing topology
00:49:47 - Headlights
00:51:02 - The Knife tool
00:53:20 - The truck roof
00:55:47 - Denoising the highpoly mesh via Blender sculpting
01:02:21 - Export to Reality Capture
01:05:38 - Bake textures in Reality Capture
01:09:07 - Creating materials in Blender
01:14:12 - Fixing the roof (clone brush)
01:17:52 - Upscaling the diffuse texture using neural network (CupScale free tool)
01:21:39 - The last minute tweaks
01:25:13 - Conclusion
Красавец, четкий туториал,всегда двигал точки (vertexes) рядом с окнами машин в ручную чтобы потом применить материал и даже не догадывался что можно просто сделать кнайф проджект🤙🤙🤙
Highest quality Blender tutorials. No rambling, advanced topics. I love Gleb!
Thank you so much! This kind of quality workshop makes me so glad to be alive at this point of time.
I was always curious about photoscanning, but after seeing all this process, I think it's still easier and less resource-consuming to just make your own model from scratch.
Depends on the subject matter, for things like tree stumps and organic shapes, photoscanning really shines. For hard surface stuff it's not the best and needs a lot of cleanup
My take on photogrammetry vs modeling from scratch is: it depends :)
I would say photogrammetry is still good for hard surface stuff when we are talking about something as destroyed or even more destroyed than this truck. It's very difficult to build a perfect hard surface model and make it look worn out or destroyed and look natural at the same time. Or maybe I just don't have the patience for that. Imagine a rusty destroyed tractor still filled with hard surfaces. But will look much better via photogrammetry than trying to achieve something only by referencing photos and texture painting.
The truck looks like it’s been at the bottom of the ocean for decades. So now I know a great way to create that. Thank you! :)
Haha, cool! :D
Рашн сабтайтлз реквайред. Джаст джоукин. Гуд джоб!
Great tutorial and the end result is awesome. But that's a loooot of work and would probably take me three times as long as actually building such an object from scratch and texturing it with the textures instead. Not saying that my end result would be as good though..
Thank for this amazing tutorial.. love from india ❤
❤️
Thank you for your video. It helps me so much for my job.
Wow! NeuralNetworks for upscaling result of photogrammetry is amazing trick! Thanks for tutorial :)
Dear Gleb & Team, thanks for make this Tutorial happened! Cheers
The model in the end doesn’t look bad. Definitely usable for something in a background. I think the one thing that bugs me about it is the glass. The waviness of it gives off a red flag that something isn’t realistic as the glass would either be in one piece and smooth, or broken and missing or with holes. I think it’s best to just remove the normal map from those objects since they would have a very uniform smoothness, and use the roughness to give some imperfections. Throw on another image and use a color ramp to define which parts have the dirt and grime.
Anthony, I agree, generally we would expect the glass to be more or less undisturbed by normal map. That being said, for some reason which I can't put my finger on it looked more believable *with* a normal map (not even with a normal map, but with an EXTRA large-scale noise applied to object normals - as seen in the project files). At least, that's my perception of it. In one of the test I even exported a broken mesh from RC (it has 'fixed' geometry by making it manifold, which turned the one-sided objects into a self-intersected mess). And this broken self-intersected geometry looked fairly believable when assigned the glass shader, maybe due to fake depth effect introduced by mangled normals. There's something to it, I just don't know how to describe it properly I guess :)
Great addition for the awesome photogrammetry course Gleb! 🤟🏻
We felt that the 'smartphone capture and then fixing the mess' workflow didn't get the attention it deserved, so this is our attempt to fix it somewhat. :)
really great tutorial, thanks!
Is there any way in Blender to reproject the vertex color between different models? Thanks for the great tutorial.
24:06 what I like to do is select what I want to keep with select linked (hover over the part and press L), then invert the selection (ctrl i), and then delete.
Edit: looks like you do end up mentioning this method at 30:15 for getting a single vertex from a plane. For this, there is a built-in addon that adds a single vertex to the add mesh menu. I can't remember which addon it is though.
nice job
Greatttt tutorial
Uday, thank you kindly :)
You would save so much time by getting a cleaner scan from more pictures and angles
that knife project shit was the thing I was waiting for for 36 minutes and for the life of me I can't do it. I repeat your steps and it doesn't project or projects wherever it wants 😄 I did it the right way by chance once and can't reproduce it now for 15 mins. Some things are just not to be. Reminds me of my high school math classes.
Thx for the vid though. I'll still learn whatever I can from it 🥲
I feel your pain, Donald :) It is so frustrating when something just doesn't work.
Greetings from Warsaw 😀
Great tut. As always 🖖
I'm curious to know the specs of the rig you work on, sculpting was so smooth, I don't think my PC would be able to handle that easily if at all.
Thanks for the wonderful lesson! Can you please tell me how to make an ordinary polygonal model with the correct mesh from photogrammetry of a car model? Perhaps there are some special programs for this? I would love to watch a video on this topic.
Nice! But it's faster and cleaner to build it from scratch :)
Such an awesome tutorial!
I was wondering... For the interior... Is it just as easy to get rid of all the garbage inside and whip up some basic shapes that somewhat resemble car seats, steering wheel etc instead of leaving it as it is? Or.. Is that a whole ton of other work? You make everything look so easy and Im thinking Ill hafta watch this about 10x (or more) to "get it" .
Great tutorial, thank you.
Glad you liked the result, Matthew!
Dziękuję bardzo Gleb! Subscribed for more pro tuts!
amazing thank you so much !!!!
15:22 Nvidia GPU with CUDA 2.0 compatibility at least (just for my near-future reference Thanks!)
Can we use mesh-room instead of reality capture?
What would be a good method to create very fast trash models of the cars. The faster, the better. I would use those for our car wrappings, which I sometimes design in my job for our clients. Lidar?
I'd say, just a regular photogrammetry and then quick retopo and texture baking. That would be my first thought.
@@GlebAlexandrov yeah, I think that makes sense. My biggest worry is to save the money of the clients so that we don't need to bill them too much of work that can be done somehow faster. But this 3d-thing quite much is just an addition. The whole process obviously can be done without using any 3d softwares. It just is cool that we can (if needed or wanted) show the end result in 3d before physically touching the car. 🤔
made in Lodz ? ;) good scan nice tut.
Great effort
But i think remodelling the car takes less time and better visualization
did you ever try polycam app photo mode in iphone? it's kinda work well and fast too
i think that it's faster to make 3 photos - from side, from front anf from back - anf than model this in lopoly and than make a photo-based textures
#Gold #Blender #Photogrammetry Thank you.
Hey. Just downloaded the video onto my PC. When you make a 3D truck, how do you work on putting in the interior such as the dash, shift lever, and a control on how to activate the windshield wipers, etc.?
Judging by the license plate, this was taken in Poland, Lódź?
The master ❤️
Honestly - I think it would be easier and better to just do the retopo of this model :D Unless you REAAAALYYYYY don't want to mess around with texture projection and so on :D
I have followed the tutorial step by step. At timestamp 1:06:13, reimport back into Reality Capture I get an error, "The model contains non-textured triangles." What have I done incorrectly? (Great video by the way, I appreciate your providing it as I have been wanting to watch your Creative Shrimp photogrammetry course- just not in the budget at this time).
The project files link in the description is broken :(
Isn't it faster to use the hotkey ctrl + A + G (apply visual geometry to mesh) instead of permanent convert to mesh? Thanx for tutorial ^_^
How much time this cleanup process actually took you to complete it till the final output. Not photogrammetry just remodeling & texturing touchups?
What is the polycount? As it doesn't look like very low poly model?
Есть ли способ и смысл показать программе какие точки на фото стоит учитывать и почистить ошибочно определённые совпадения? Например, если фигура является созданной в параметрическом редакторе машиностроительной деталью и состоит из ровных плоскостей, окружностей, углов предсказуемой величины, тел вращения и тому подобного, может просто отметить на фотографиях несколько точек вручную по углам и вдоль кривых?
fire
Could it be reconstructed with pictures that are downloaded randomly from google image search? .. (then some duplicated and mirrored to represent the other side of the vehicle)
Do you mean by the means of photogrammetry? I don't think so. Otherwise, using the random pictures as references for the regular modeling so to speak, yeah possibly :)
@@GlebAlexandrov alright..it seems Autodesk Image Modeler 2009 is still the way to go
Hi Alex, could you send me the files as the link is not working
hi can you use an android phone
Nice work!
Just curious, isn't it faster and better quality to just model a lowpoly object, unwrap and use the images as textures?
I think that would be a good strategy for a hard surface model like this car. Gleb's technique would work amazingly for organic photoscans!
Yes and no. I think it would be slower to do a retopo, but final result would look much better. But it's not necessary for a BG model that will be somewhere beyound level bounds or just out of player's/viewer's reach.
That's not a way to do a main focus object tho. For that you'd have to ideally get a blueprint of a car model it with all the details you can see. Or at least do a proper retopology + model in a few details like shock absorbers axles etc.
hi the project files links are'nt working
wiem że jesteś polakiem bo tablice z elbląga na busie to napisze po polsku, mordo link przestał działać do pobrania plików projektu jeśli by sie dało to jakoś odnowić czy coś byłbym bardzo wdzięczny bo chciałem ozbaczyć czy mój procesor sobie wgl poradzi z odpaleniem takiego projektu w blenderze
Could you do this using iphone 12 pro LiDar camera?
LiDar is not a camera
😂😂😂 I think he’ll still understand the question… although he has ignored me so perhaps he doesn’t… 😂😂😭
Я и так не меньше литра выпиваю в сутки. Спасибо за урок.
hmmm ... looks like you can model this truck from scratch by using reference images for such amount of time.. what you may need that bad mesh for? a cinema VFX prop? Looks completely useless for gamedev (for hard surfaces tho). Rocks and other natural objects i think will be good enough.. but cars? idk...
A background asset most likely, I'd say. At least that's my perception of it. Out of curiosity, how much time you think it would take to model something similar from scratch, based on reference photos? I assume you mean the photos from internet, or do you mean front-side-etc projections?
yeah, I mean Front\Side\Rear view projections and not very detailed result where you can simply use your photos as basic textures. You will end up with smth close to the result of "3d scanning" (or however it' called)) but more versatile in terms of use. For example that wheels I suppose are easier and faster to be modelled from cylinders than cleaned up on a 3d scan)
@@GlebAlexandrov you could block that out in moments and add detail in a second past far faster than your method. Although very interesting nonetheless
ЧТОТО БОЛЬНО МНОГО ЛИШНИХ ПРОГРАММ - Написал бы хоть название
Waait, you live in Łódź? :D
You were in Łódź?
In Warsaw, but I'm sure it's possible to find a similar truck in Łódź too)
@@GlebAlexandrov the license plate confused me :)
@@mariuszurbaniak7314 ah, right! you have a keen eye)
First comment
how do you manage to be THAT fast? )
what phone takes raw format photos? with that you lost me already...not sure it wouldnt be faster and cleaner to just 3d model something and sub painter it and be done with it. this is not really clean enough for a game. and takes almost the same amt of work or more than a more traditional approach. seems like this process is not quite there yet.
I feel bad for losing you. Some phones support RAW out of the box, some need an app to shoot in raw (correct me if I'm wrong though). Open Camera app for Android for example (not all cameras support it still).
Thank you, for extensive tutorial. I need to say what it can be shorter. Too many details you can skip and get to main points much faster.
This is finally a tutorial that covers almost everything. Skipping over details is what would make me not understand certain parts. You can always speed up if you want and skip certain parts
alexandroff))
Łódź?
If this is low poly, then Tomb Raider 1 graphics was HDR 8K quality
Would be easier to just model it.
какой-то у него восточно-европейский акцент
So russia
I know what you mean, it's that kind of vibe for sure. It's Poland though, but let's say, it's in the same Half-life 2-esque universe when it comes to abandoned trucks :)
@@GlebAlexandrov Greetings from Kaliningrad! Bardzo fajnu kontent tu masz!
@@denischikita Dzięki! Стараемся :)
Just too complicated.
Глеб видел что твориться в Мариуполе? Российские войска бомбят жилые кварталы