[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vr / w / wg] [i / ic] [r9k] [s4s] [vip] [cm / hm / lgbt / y] [3 / aco / adv / an / asp / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / qst / sci / soc / sp / tg / toy / trv / tv / vp / wsg / wsr / x] [Settings] [Home]
Board
Settings Home
/3/ - 3DCG



Thread archived.
You cannot reply anymore.



File: 1497491531368.jpg (63 KB, 720x540)
63 KB
63 KB JPG
Relatively noob 3d fag here, gradually becoming serviceable in 3ds and new to the board. I discovered photogrammetry while doing research for VR shit, and am hunting for Android apps that aid in producing accurate 3d models; hopefully a sort of inverse photosphere layout that might aid in angle and distance consistency. I use ReMake for processing and have had some success with DSLRs, though I cannot pic related it at the moment.

Also photogrammetry general, pic not related
>>
You're going to lose interest in 3D scanning as soon as you realize that it doesn't actually work.
>>
>>568509
I've actually been pretty thrilled with some of the results I've gotten. Don't know what I'd ever do with them, but it's a fun novelty
>>
>>568509
But it does and it is beautiful.
>>
File: flightcomp3.jpg (69 KB, 500x536)
69 KB
69 KB JPG
>>568523
I got my first positive result using pic related as a base for my target object (a small plastic wolf) and circling to the best of my estimation with a DSLR. Unfortunately I was shooting prime and it turns out the diopter wasn't set quite right, so the imageset was collectively focused a little too far. The result was still impressive, managing to capture even the texture of the fur, but some if the harder lines behaved poorly and fused part of the mouth and tail. Very encouraging result though. It seems to be the base that made the difference, as my previous attempts ranged between terrible and completely nonsensical. I've heard newspaper works fairly well, but I haven't tried it yet.
>>
Here's an awesome video with Alex Alvarez talking about this topic, if anyobody is interested. Useful info on where to start, too.

https://www.thegnomonworkshop.com/blog/an-evening-with-alex-alvarez-creating-natural-3d-environments
>>
>>568605
Learn how to model with subdivision surfaces. Learn about mesh topology. You'll understand why 3D scanning doesn't work.
>>
>>568509
>You're going to lose interest in 3D scanning as soon as you realize that it doesn't actually work.

I have a crown in my mouth that feels 100% like the tooth it replaced. Its underside 100% matches the point he ground the original tooth to. The crown was milled based off of 3D data the dentist took in the office.

> laser scanner
Nope
> complex scanning apparatus
Nope

The entirety of the meshes were taken from digital PHOTOGRAPHS with absolutely no reference markers placed in my mouth. The software is THAT good at not only stitching the photos together but inferring the shapes it took pictures of and building a cap mesh, a point mesh, boolean subtracting the latter from the former and creating a mesh so high res that the crown milling station produced a completely smooth, perfect fitting duplicate of the tooth that was in my mouth an hour prior.

It wasn't the camera (a tiny CCD device on a stick that is probably not much different from the ones in mobile phones), it was the software.
>>
>>568612
Are you sure it was 3D scanning/printing? You seem completely lost in the hype. My dentist does that by taking a mold.
>>
>>568609
What the fuck are you even talking about? What "doesn't work"?
>>
>>568609

>>it doesn't work

Erm.... You may want to check out the work of a little Swedish company called Quixel, specifically their 'Megascans' products. Try telling them that it doesn't work.
>>
>>568609

>>>3D scanning doesn't work

Erm......... What the fuck do you call this then?

https://youtu.be/Eex8hzWWKhM
>>
>>568625
>>568628
That's what I was also thinking. Btw, I've been trying some of these out for the past few days now. Shit is incredible. And as I said, check out Alex Alvarez talking about how he scans his assets, they work perfectly as non-hero objects, but could even work as the main subject. While you are talking how 3D scanning doesn't work, I'm wondering why are we still even modeling some things when photogrammetry exists.

I'd also like to try it but first I need a proper DSLR.
>>
>>568628
Useless junk?
>>
>>568632
Hi Alex.
>>
>>568632
Honestly phone pics would probably be fine at modern megapixel numbers, but it would be a big help to have a guide as I suggested in the OP. I have to imagine such a thing exists, but I can't find one for the life of me

>Why bother modelling
Having tried and researched both, I don't see a good way to use photogrammetry for non-natural props. Trees, rocks, and other irregular objects can be made to look amazing, but regular objects tend to come out with wavey, uneven sides where they ought to be flat, mine or otherwise. They look great textured and lit, but don't stand up to scrutiny. Definitely still trying to get good, but photogrammetry could offer some terrific scatter and natural features
>>
Btw the next iPhone is rumoured to have some extra hardware in the camera module which is going to make photogrammetry apps work 1000 times better. I don't know what that could be exactly - maybe just twin lenses, maybe some sort of range/distance sensor, I dunno.
Make of that what you will.
But whatever, it may be the case that, yeah, if you want a little prop of a real-world object and you aren't gonna need it too near the camera, you can totally just squeeze it out of your iPhone in 2 minutes.

It could be interesting, if nothing else.
>>
>>568690
Is it really possible to create anything decent with a smartphone camera? I heard that you need to have a really good DSLR to get proper details. I've seen some of those apps, but they seemed like something just for fun, since the models looked pretty shitty, but then again, it was years ago.
>>
>>568693
Ive never looked into photogrammetry on the iPhone but Fwiw I've lost count of the amount of things I've textured using photos I've taken on my phone and iPad it's totally fine for that stuff so I don't see why it wouldn't be at least adequate for photogrammetry of small objects that would appear in middle distance in whatever you're working on.

That said - getting good textures and photogrammetry largely comes down to lighting. If your subject is unevenly lit, or under lit, or has several different light colours hitting it, it doesn't matter what camera you're using - it won't look good.
>>
>>568693

Leaving aside the lens and sensor hardware on a smartphone camera, a smartphone has other pieces of hardware that a DSLR does not which would help with photogrammetry - the accelerometer being the most obvious.

Yes, everyone has seen the useless novelty crap that their kid brother produced with some free photogrammetry app - but mark my words, in the next 18-24 months, you're gonna start seeing some seriously good looking meshes which have been made just with a smartphone.
>>
>>568690

Yes - Apple seems to have VR and AR firmly in its sights all of a sudden. Tim Cook has mentioned it a few times recently, this new crazy hi powered iMac they're making is aimed at VR content producers, so it would make sense that their new phones will have additional spacial sensing features.
>>
Are you done talking to yourself yet?
>>
>>568609
what is retopo?
>>
File: copythat.png (94 KB, 337x456)
94 KB
94 KB PNG
>>568690
>tfw 3DCG phones
>>
>>568693
I've been fiddling with scann3d while out of town. The results haven't been half bad, but the fidelity is fairly obviously limited. I'm going to run the same imagesets through ReMake when I get home, so provided this thread is still up tonight or tomorrow I'll post side-by-sides.

Pic related, a rock. I was fairly close so the mesh is acceptably detailed, but it doesn't blow me away just yet.
>>
>>568730
Update: Here's the same imageset at roughly the same angle, but this time compiled by ReMake. I'm pretty thrilled.

For reference, the model was compiled, by both the phone software and ReMake, from the same 44 images taken with a Nexus 6p. The future might just be now.

>>568609
>learn how to model with subdivision surfaces/about mesh topology
Any good resources I should go after? Trying to get gud; progress has been slow but rewarding.
>>
>>568819
Damn, that looks pretty good. What phones would be the best for this? I need to buy a new one soon anyway.
>>
>>568826
I've been consistently amazed by the Nexus series cameras; I can't speak for many others though. Photogrammetry seems to gain the most from deep depths of field (high F), and consistent focus.
>>
>>568826
Wait for the new iPhone.
>>
Please post more of your photogrammetry. Your pictures af an height field with a random rock textured on top are the best I've ever seen.
>>
>>568826
>>568605
Here's the out-of-focus wolf, for the curious. I'm doing another of him now, this time with the phone camera. Going to try both a scann3d compile and a remake compile again. While my phone is working I'll task ReMake with a rock fountain walkaround I did and compare results on those too.

>>568878
I'm pretty happy with the results, though I haven't done any finish work to make it useful. Not here to brag, just investigating a technical curiosity.
I could post some of the source images if desired. This being 4chan though, I don't have a good way of offering an actual model 360. You'll just have to trust me on that.
>>
>>568884
Looks not creepy at all. I'm glad that all that expensive hardware is in your hands.
>>
File: RockPhotogrammetryComp.png (2.7 MB, 1700x1054)
2.7 MB
2.7 MB PNG
ReMake on the left, Scann3d on the Right, both compiled from the same set of 23 photos. The ReMake solution seems surprisingly good given the lighting conditions were pretty awful.

The quality cline between ReMake and Scann3d here is a bit sharper, presumably due to the latter wasting a lot of what appears to be a fairly low, hard polygon limit on non-target objects. I'm not as impressed by this one as I am the rock, probably due to poorer lighting and technique.

The second phone wolf attempt failed as Scann3d blew most of its polygons on background again. It really seems to require close, even macro positioning to do well, and my phone just doesn't focus closely enough for small objects. That photoset is going into ReMake next.
>>568884
>creepy
Yeah, he's got kind of a silent hill look to him there.
>>
>>568901
That thing must have hurt a lot when it came out.
>>
>>568615
he literally put a wand with a camera in the tip into my mouth and took a set of photos that were reconstructed into a color 3D model he could turn in realtime and the software recognized by color and texture what was tooth and what was gum. I shit you not.
>>
He put a wand into your mouth.
>>
The wolf didn't go well, but enough happened to see the difference proper focus made. Small objects are hard, it seems.
>>
>>568927
Not enough photos at the right angles it looks like to me.
>>
>>568509
its used for fx
>>
Ive tried agisoft and it was good, stand alone is 170 tho, I say avoid mobile phone gimmicks, but Im unsure if youre a photographer (I assumed but im probably wrong)

if you have a decent camera you can get super detailed models, but getting textures into an optimized geo is another story
>>
>>568914
I had this kind of work done too. I refuse to get any other type of filling or crown now after that.
>>
ITT: Traditional modellers feeling threatened by technology making them obsolete
>>
>>568655

You make it sound like the scanning tech can't be honed. Of course these early iterations will be obsolete compared to professionally made 3D models. Doesn't mean they won't be good enough to scan and retopo all kinds of shit on the fly in a decade or two.
>>
So what's the most no-fuss free way of doing this stuff? I really liked the environments in SteamHome and want to try making a few myself. Even heard you can rip 360 pics from Google Streetview and they'd work.
>>
>>571244
Here's something I wrote back in 2014. The situation might've changed, I haven't checked on progress yet. If the links are not dead this should work otherwise.
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
I've made my little investigation for free solutions for model, texture creation through photogrammetry.

The current it tool is VisualSFM: http://ccwu.me/vsfm/
It uses SIFT for feature matching to then solve camera position. After that it can use CMVS/PMVS2, CMP-MVS or SURE for dense point cloud reconstruction.
Point cloud data can be imported into Meshlab to create mesh and texture. Mesh can be exported to obj. along with texture.

The workflow using CMVS/PMVS2 is explained in this article: http://www.academia.edu/3649828/Generating_a_Photogrammetric_model_using_VisualSFM_and_post-processing_with_Meshlab
Video explanation for the workflow: https://www.youtube.com/watch?v=Ax6gajFE-qw
Guide to installing VisualSFM: http://ccwu.me/vsfm/install.html
VSFM feature explanation: http://ccwu.me/vsfm/doc.html
Meshlab: http://meshlab.sourceforge.net/

Potentially better results can be achieved using CMP-MVS.
It requires a CUDA gpu. Because emulating doesn't work I can't confirm how comparatively good the results are.
Video explanation for the workflow: https://www.youtube.com/watch?v=Yt9MmQHobTI
2nd (newer) Video explanation for the workflow: https://www.youtube.com/watch?v=ZRTEMKS3Sw0
CMPMVS: http://ptak.felk.cvut.cz/sfmservice/websfm.pl?menu=cmpmvs
CloudCompare: http://www.danielgm.net/cc/

Photo arrays for testing can be downloaded from Microsoft's Photosynth website using Photosynth Toolkit: http://www.visual-experiments.com/demos/photosynthtoolkit
Photogrammetry presentation, long: http://video-jsoe.ucsd.edu/asx/AguerayArcas.asx
Short: http://www.youtube.com/watch?v=6cnP8FqRoPI




Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.