Jump to content
Search In
  • More options...
Find results that contain...
Find results in...
Fix

3D models from photos of the miniatures ?

Recommended Posts

Hello,
I've been looking but haven't found anything linked to that question yet. I'm looking for 3D scans or reconstructed models from photos, such as ones made freely by 123D catch.
Or I'm looking of photos sets of these miniatures to rebuild them in 3D myself with some more powerfull tools (such as Recap photo or Photoscan).
My goal is to capture them, remove the useless parts (such as supports for flying monsters), project the sprites colors on them and then share the cleaned result to the community .
This would be a base for making real time 3D version of these, or 3D print the miniatures.
My own project is to re-use them as a strong solid start to reproduce the sprites in 3D as closely as possible, then generate voxels from these, because I believe in voxel doom. I already started some experiments on the items with polytovox, and the results are promising.
Other benefits of this conversion would be the ability to add interpolated animation frames to the actual sprites, or add other animations, matching the vanilla monsters as much as possible.

I doubt somebody did the scans yet, but if some of you have the miniatures and a digital camera, maybe you could take and share the photos of these ?

Note 1 : I'm not sure it will be more efficient (fidelity wise) than directly rebuilding the sprite in 3D.
Note 2 : I already tried to submit the cyberdemon sprite's 8 angles to recap-photo or 123d-catch, both haven't been able to rebuild the monster in 3D, that would have been the best first step.
Note 3 : I'm also starting to look for 3D meshes/animations that already exist, that are very close to the vanilla sprites. Why redo the work if there is already top quality stuff out there :)

thanks !

Share this post


Link to post

Sadly, the thing you desire does not already exist.

I'm in the middle of making my own versions of the Doom monsters but I have taken a lot of creative liberties in my versions.

Share this post


Link to post

This model looks fine, but I'm indeed looking for vanilla things.

I've found on ebay some multiple shots of painted miniatures, tried to feed the recap photo with thoses, but 4 pictures is really not enough for this software.

Share this post


Link to post

For one, I don't believe you can get results as easily and cleanly as shown in that 123D Catch commercial: there are several constraints on the complexity of what is reconstructable from a finite number of photos and angles, and elements such as holes, concavities etc. might be troublesome. You might end up doing tons of manual image editing and later on, 3D model cleaning anyway, to the point of it being just as much work as modelling by hand. If you factor in animation and different poses, things will get even uglier.

And that's assuming that you can find the original id models. I think only the Baron, Archvile, Cyberdemon, Spider Mastermind, Demon and Revenant (and the Player?) had models, with the rest being hand-drawn at their in-game resolution, so they never were "hi res" to begin with.

As for making voxel models out of them...sounds like a memory-killing app, and nearly impossible to animate at the current state of the art, without ending up resorting to polygon meshes/skeletons once again.

Share this post


Link to post

A while back I took photos of one of my imp figures from 8 angles with the idea of turning it into in-game sprites.





That's as far as I got, though. This probably doesn't help build models, unfortunately.

Share this post


Link to post
fraggle said:

A while back I took photos of one of my imp figures from 8 angles with the idea of turning it into in-game sprites.


WTF @ size:



Did it come with a lens in order to be able to admire it? O_o

fraggle said:

This probably doesn't help build models, unfortunately.


well, there's software that claims to do exactly that from a sufficient number of photos. In order to produce animation though, you'd need a reposable model, or supply an animation skeleton and segment the body into rigid parts, joints etc. in order for it not to look like ass. In other words, it would take just as much work as building it from scratch, perhaps basing it on a humanoid template. No software will allow you to progress beyond the point of making ONE single-pose model if you don't already have advanced 3D modelling skills to handle the necessary animation, re-texturing etc.

And even supposing that one monster IS finally complete, where would one use it? If the answer is "I'd take screenshots at 8 angles and convert it to a (high resolution) sprite" then don't bother :-)

Share this post


Link to post
fraggle said:

A while back I took photos of one of my imp figures from 8 angles with the idea of turning it into in-game sprites.

That's as far as I got, though. This probably doesn't help build models, unfortunately.


Thank you very much !
I'll experiments with these as well. I think the specular highlights of the metal will bug the reconstruction, but I'm really curious and eager to learn.
Thank you, I'll use the raw images !

But I think I will have to remodel the sprites, most of the miniatures have already too much differences to the original designs.

Share this post


Link to post
NiuHaka said:

Sadly, the thing you desire does not already exist.

I'm in the middle of making my own versions of the Doom monsters but I have taken a lot of creative liberties in my versions.

IMAGE


HOLY SHIT MAN! I can barely contain my excitement!

Share this post


Link to post
Fix said:

Thank you very much !
I'll experiments with these as well. I think the specular highlights of the metal will bug the reconstruction, but I'm really curious and eager to learn.
Thank you, I'll use the raw images !

But I think I will have to remodel the sprites, most of the miniatures have already too much differences to the original designs.


Whew! I really hope you don't plan to learn this for a career. It would be a lot of learning into something you can't really use in the real world any more in respect to sprite generation.

DooM_RO said:

HOLY SHIT MAN! I can barely contain my excitement!


Hehehe! Oh, just you wait, it only gets better!

Share this post


Link to post

Using a laser model scanner you could create a voxel model, of course you would need access to one. If you know any friends who are in that kind of business or who have access to certain medical machines they may be able to scan the models.

NiuHaka said:

Sadly, the thing you desire does not already exist.

I'm in the middle of making my own versions of the Doom monsters but I have taken a lot of creative liberties in my versions.


Looks very nice.

Share this post


Link to post
NiuHaka said:

Whew! I really hope you don't plan to learn this for a career. It would be a lot of learning into something you can't really use in the real world any more in respect to sprite generation.


I already have a career (thanks doom for showing me the light 20 years ago :) ); I am a strong believer of photogrammetry, since years.
Anyway, the sprite / voxel project is more for fun (point clouds are not that far from voxels).
I'll learn very usefull things from this experience, the sprite generation is just a little part of it; and hopefully will come back with interesting things for this community.

update : I've found a video showing multiple angles of painted miniatures.
Processing the demon right now =).

Share this post


Link to post
Maes said:

Did it come with a lens in order to be able to admire it? O_o

The smaller figures are all that size, I guess so they're at the right scale to the larger monsters.

well, there's software that claims to do exactly that from a sufficient number of photos.

Yeah, I think you'd need more than 8 figures, probably you'd want at least 100 to be able to build anything resembling an accurate model. And you'd have to get the angles precise, too - meaning you'd need some kind of turntable able to turn exact angles.

In order to produce animation though, you'd need a reposable model, or supply an animation skeleton and segment the body into rigid parts, joints etc. in order for it not to look like ass. In other words, it would take just as much work as building it from scratch, perhaps basing it on a humanoid template. No software will allow you to progress beyond the point of making ONE single-pose model if you don't already have advanced 3D modelling skills to handle the necessary animation, re-texturing etc.

The original Doom models were built as clay on top of wooden art manikins which I guess made the reposing part easy. I don't know much about 3D modelling software but maybe after you've built up a model and added an underlying skeleton it's easier to repose? Skeletal animation is pretty well-developed nowadays.

Share this post


Link to post
fraggle said:

Yeah, I think you'd need more than 8 figures, probably you'd want at least 100 to be able to build anything resembling an accurate model. And you'd have to get the angles precise, too - meaning you'd need some kind of turntable able to turn exact angles.

Yes, 100 is a good amount, but the quantity isn't enough.
The turn table (used in the video I found) is a wrong choice, because the lighting on the model will change on each photo, the best is to turn around the subject and have the most stable conditions of lighting as possible.

fraggle said:

... but maybe after you've built up a model and added an underlying skeleton it's easier to repose? Skeletal animation is pretty well-developed nowadays.


Yes it's the idea, transform the mesh into a traditional character setup.

Share this post


Link to post
fraggle said:

The original Doom models were built as clay on top of wooden art manikins which I guess made the reposing part easy.

that's news to me. I'm looking at the proportions of the doomguy and an average mannequin ( here) and there is no way that the clay was just plastered on top of it. Shoulders are not wide enough, nor are the hips.

edit: I've been looking around and, proportionally, the Quake 3 doomguy isn't actually a bad match. And animating it to match the original walk cycle would not be hard. If you were going to do a proof of concept, then that is where I would start.

The problem you are going to have converting polygons to voxels is a low voxel count. Even assuming that there is a way to convert a skinned polygon model into a coloured voxel model (i'm sure there is, but the tech is beyond me), individual voxels will anti-alias badly. You'd have to recolour each individual voxel on each indivdual animation frame by hand.

I'm not saying it's impossible, but it's something you need to be aware of. I'd reskin the Quake 3 model and work from there; I might be inclined to make a basic, original doom pallette voxel-friendly skin if you can get a basic proof-of-concept going.

Share this post


Link to post
darknation said:



yeah, apparently you can convert mesh + skin into voxels. No one was more surprised than I was.


Holy shit this is amazing

Share this post


Link to post

WOW!

Would be cool if this could be done with textures as well to give them more depth. TEKWALL4 would look amazing if remade like that.

Share this post


Link to post

Woahoooh! Awesome!
I don't know much about voxels. Would the voxel have a fixed "pixel" size? Are in game HD voxels possible?

Share this post


Link to post

Yeah, but a single voxel item is limited to fitting in a 256-sided cube (at least for KVX format; I'm sure there are other voxel formats with larger limits but it's KVX that's supported in ZDoom and DelphiDoom).

So a high-res voxel is a tiny voxel.

Share this post


Link to post
darknation said:

yeah, apparently you can convert mesh + skin into voxels. No one was more surprised than I was.


Whoever said that you can't? The problem is that, just like producing 2D sprites from models (either virtual or real) or vector drawings, you have to degrade a resource of practically infinite detail to a quantized version of itself, with a finite number of pixels (or voxels). In other words, you turn your smooth drawing or smooth model into a bunch of blocky pixels (or voxels, if we're talking 3D). No matter how arbitrarily high you set the export resolution, there's ALWAYS going to be a scale where it looks like ass.

Plus, with voxels, you lose the ability to do real-time skeleton-based animation and other vector transformations that are possible with traditional polygon models (name me ONE current game engine that can do it), so you basically end up with a 3D version of 2D sprites, with all of their shortcomings, and memory requirements blown up an order of magnitude. Ironically, nearly all workarounds to those problems include re-converting voxels into polygons (!), essentially making voxels a bunch of cubes, making the worst possible use of both polygon and voxel technology (unless we're talking about Minecraft or Lego games...)

Share this post


Link to post

First ugly Experiment done + mockup of ingame result.

The mesh is too different from the original design.
A manual remodeling will be more accurate as expected, but the miniature reconstruction will certainly speed up the process.

Share this post


Link to post

Just out of curiosity, how long did it take you to produce that one frame, even by having a ready made, full-quality 3D model at your disposal, and without manually correcting anything?

Add model capturing time, post-processing, etc. and multiply it by number of monsters and number of different poses (at least rotations come "for free").

Share this post


Link to post

I did the first conversion experiment in 3 hours I think, after having received the mesh. But I tried to project the original sprite as a texture.

The final conversion into a voxel is very fast, a few seconds I'd say.

-Export the mesh as .obj, use poly2vox.exe to convert the .obj into a .vox.

-taking the photos and then start the upload to Recap360, a few minutes.

-processing of the photos, hours but you just have to wait for the result, so this doens't count

-cleaning the result : can be very long. I think Infinite-Realities process to clean human 7000k scans is one day or more. But here we're aiming a very low res, so I'd say 1 or 2 hour max before having a workable mesh.
I already did some real life scans almost ready to use after openning, here the video from youtube as a source wasn't very good.

Share this post


Link to post
darknation said:

that's news to me. I'm looking at the proportions of the doomguy and an average mannequin ( here) and there is no way that the clay was just plastered on top of it. Shoulders are not wide enough, nor are the hips.

I'm certainly no expert, but you can check out the photos of the Baron/Player figures on John Romero's Facebook page and it definitely seems that they had art mannequins for skeletons (visible because they've both suffered damage over years).

Fix said:

First ugly Experiment done + mockup of ingame result.

This is awesome!

The 123D Catch system looks really interesting. It's unfortunate (but understandable) that it won't work with reflective surfaces. I have a full set of the miniatures and would offer to take pictures, but they're as-yet unpainted so it's unlikely to work. But I can give it a try if you want.

Share this post


Link to post

That voxel Demon is so badass looking. I wanna see an animated version of that thing sooooooooo badly.

Share this post


Link to post
NiuHaka said:

That voxel Demon is so badass looking. I wanna see an animated version of that thing sooooooooo badly.


And here's the catch: are there more miniatures of the same Demon in different poses? If not, then you need the skills of a traditional 3D modeler and animator to take over from that point onward, so from that point of view, this method does not provide a shortcut that takes you from pictures straight to a fully animated 3D model. Let alone that ports with 3D models employ skeletal animation, so a simple sequence static frames/poses wouldn't do it.

Also keep in mind that for other poses, certain details of the model might need to be drastically changed, e.g. muscle definition, head, eye and mouth position etc. amounting to, once again, a full-time job.

Perhaps the job of a modeler would be made easier if he had different "statuses" of the same monster (e.g. making motion tweening/animation easier).

Share this post


Link to post
fraggle said:

The 123D Catch system looks really interesting. It's unfortunate (but understandable) that it won't work with reflective surfaces. I have a full set of the miniatures and would offer to take pictures, but they're as-yet unpainted so it's unlikely to work. But I can give it a try if you want.


If you have a soft lighting environment (indirect diffuse light : aim lights at walls and ceiling, or cloudy outside), that could help minimize reflections.
I think it is worth the try anyway. Maybe it will be good enough to work with the unpainted miniatures.
Thank you very much for your offer, this is really great !

Some instructions to help the software:
-place the miniature on a non repetitive/tiling surface, a book cover or newspaper is good.
-sharp images, minimize depth of field (hard with miniatures :)
-stable lighting (no flash), exposure, balance, focus, etc...
-I'd start with a 360° on 3 angles :
----30°bottom (if possible ! (use of a mirror ?)
----front
----45°top.
-Recap photos can take up to 250 pictures, but I know it's really tedious to take so many photos (extracting them from a movie is easier, but the quality of the extracted pictures is not as good, and we loose the informations stored in the photos).
So I suggest 10 to 20 photos per 360°. they say it's better to have a few very good photos than a lot of bas ones.

If you have an i-phone, you can try the 123Dcatch application directly with your phone (doesn't give results as good as recap photo though)

Maes said:

amounting to, once again, a full-time job.


I agree with you and never fooled myself on that part.
I've been making game assets since doom, so I know this is a lot of work.
However, I think the size of the sprites, the number of frames makes it a bit lighter than next gen production.

Share this post


Link to post

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×