1:30 "After a half-hour you should see this thing" For me this part took _2 and a half hours to complete_ *And the rest of the photogrammetry took almost 6 hours* But still great tutorial, and it actually is REALLY addicting
Try 3DF ZEPHYR. It has a free version also which can do 50 images. Meshroom with 107 pics took 8 hrs on my pc while zephyr took 1 and a half. The result was also way better. Try it out, you might like it.
Virgin 3D scanning: going outside, taking a hell lot of pictures, need hours to render, too many vertices Chad Ian Hubert: just make a cylinder and extrude, add a texture and voilà
true Ian Hubert's tutorials are incredibly easy and useful, but if you want realistic 3d props or models for game scenes, photo-scanned objects are better bc of the details and realism.
I do a lot of modelling and animating, but I also got into 3D printing. So this video doubles as a way to 3D scan an object and reprint it as a model. I appreciate it!
Instead of taking photos, I took a video with the same motion and locked settings on my phone. I later edited unnecessary stuff out of the video... Then I REALLY decreased the frame rate because bro, too many frames. Output your video as a JPEG sequence and I suppose you know where to go from here!
Yup that's the only way I've tried it, but now I'm debating switching to photos like everybody else because I wonder if it outputs sharper image textures!
Coltyn Stone-Lamontagne True.... a square marker, a few strategic photos, fSpy, project from view, bake? This one is just for teaching, anyway, I guess..
or maybe you suck at modeling and creating UV textures, and this is your way to go! also there are some retopology tools out there so you can have quads and decent render times. :D
4 года назад
I appreciate how you start with a super basic photogrammetry tutorial and follow up with pretty advanced Blender show :D
Before starting this tutorial, I had never touched Blender. Nevertheless, I was interested enough to push through and scan my first object. I'm happy to say that I now have a really strange looking cream soda can that I may drop into UE4. Thanks for sharing your knowledge, Peter!
You don't photoscan a can! It's much easier to model/find a can. Photoscanning is better for organic or rough objects that would look worse if you made them from scratch.
Wow I was really bummed when I saw the display.land app was closed recently. I’m so glad you made this! It actually seems to be a lot easier than I thought it would be haha. Thanks so much Peter!!
Oh wow. I’ve fallen far enough down the Blender RU-vid rabbit hole to arrive at this video and think “I know this voice...oh! It’s Peter from Corridor!” Happy to be here and learn more for my journey through Blender!
hey man, I find taking a slow video tracking around your chosen object, then using some software that extracts as many frames as you like per second is a neat trick. it's faster than taking pictures. I find it often picks ups more detail because of the smoother camera pan around the object.
You’re amazing!! One of the best tutorials I’ve ever taken! Exactly what I was looking for. I can’t tell you how much you’ve helped me. You earned my subscription
I think you could cut some time out of this process by doing an edge split after cutting the model (vertex group may still be good for safety). That way when you select the model and ctrl-L to select linked parts it will be disconnected from the garbage.
If you have access to a DSLR and have some knowledge about it; you might want to try a CPL filter on the lens. These are made to block out 99% of reflection if you use them right
Also, any images that aren't going into the color socket should be set to "non-color data", so I usually use two copies of the image, one set to srgb and one to non-color :)
Its not an accurate height map anyway... For converting an actual height map into a normal map, you're absolutely right. But this is some arbitrary detail stolen from a colorimage... It's physically wrong anyway, so you cant mess up realistic details..... Just choose what looks better. And Ill tell you, for those color to height conversions it won't 100% always be distance. Especially flat objects that are very colorful dont want to protect the distance. The distance on those heightmap conversions comes from dark/light colors in the diffuse.... They dont resemble distance at all. They dont indicate how far something should appear when you convert them from color... They only provide some microdetail at convenient spots. By using strength (and therefore destroying the distance accuracy) you'll effectively lower the impact of dark regions next to light regions while still keeping the details on both intact. So the wrong height information from different colors gets destroyed while it will still give you the microdetails.... In this case probably exactly what you want. I discovered this on some procedural asphalt that had much lighter asphalt patches on it.... I took height from color and all the light patches looked like they were raised higher - the color tells it to do that. But by putting strength to 1 and lowering the distance to near 0 all those broad height informations got destroyed entirely only leaving the even pore details behind.
@@888marcinb Its a white hydrant with dark rust on it! It will not look better to adjust distance on this. the distance is absolutely wrong for contrastful colorimages. It would bump the rust 2cm in until some microdetail reaches the flat pieces. Adjust strength and it will uniformly transfer microdetail onto both the rust and the flat piece without creating a border around the transition.
If you're just using it for modeling and texture reference, grab an iphone 12pro and 3dscan the same object with that. While you're waiting for the photogrammetry to calculate, you can Probably model the object with decent topology and have it unwrapped by the time Meshroom is finished. Once you get the nicer data, you can selectively merge those. Like, the textures will be much better there. But the UVs suck, so bake those textures onto your new model? You can also grab some of the smaller details that didn't scan with your phone in your crappy, but still very impressive, real-time scan. Seriously, if you don't have one, find someone with one. Get the "3d scanner app" (such a terrible name) and do a scan. If you're Super anti apple, I believe the galaxy note 20, and Maybe the note 10?, has the same kind of liDar sensor. Oddly though, for smaller details, the FaceID camera has better quality, but is limited to small objects. I'm looking into seeing if it's worth it to use both.
Gl hf everyone getting into this, meshroom requires a Nvidia GPU with CUDA 2.0 or greater. You can try to customize the workflow to run through CPU only, but I warn you now it is a lot of work and totally worth it but get ready for a long journey.
I see some plants in your photo collage at the end when you're telling us to see your patreon. I'm interested to see how you were able to scan those since there are so many overlapping and thin objects. I've only tried photoscanning a few times with varying success, but that's one that I never was able to get. Any time I had occluding objects, it seemed to just get one or the other. Tips and tricks for thin and occluding objects would be a sweet video.
