next »

[FARTRON]

The old layout was already hilariously too wide making any large blocks of text very difficult to read.

This new width is even stupider.

[SirNiko]

I kind of like the new wider version.

[mono]

now i can do this

[Impmaster]

My screen isn't 1080, cause im on a laptop, so that just has a scrollbar.

[Slader16]

Oh cool, it's back to normal. Thanks Derek! 

[gimymblert]

The yulluminati were real after all

[gimymblert]

Oh, I see what you're talking about ... that's pretty far from what the kinect is doing though, but yeah, pretty cool stuff.  In my experience it's not that great in terms of the reconstruction quality ... when I tried it out, I found the results were pretty bad / it require a ton of very closely-spaced photos to really work at all.  Also some things are very hard to reconstruct w/ cameras -- a lot depends on how reliably one can extract dense correspondences, which is harder for some shapes than others.  Still, if you're patient and take a ton of photos you can probably get some nice stuff out ... let me know how it goes!

Photogrammetry test

click image and select view image or open in new tab or take the url to see bigger images





time around 15:16 taken the monday 14 july 2014 carelessly (18 photos)
Process took less than 1h (20mn cloud computing on 123D autodesk free on web) without hurry and plenty distractions
with an olympus stylus vg 180 with settings: 3Mp (1980x1440) , 4:3 , iso 100 , wb auto , compression normal, no flash
69€ locally in the west indies (martinique)
http://www.amazon.com/Olympus-Stylus-VG-180-Megapixel-Compact/dp/B00C7NX884



IMO it's cheap and indie friendly enough for production at worst as a guide, at best you need to touch it a bit. Coupled with cheap material acquisition there: http://www.polycount.com/forum/showthread.php?p=2089327 It's a win.
Practice surely will help have better result, but it's fast and and great for at least reference, not a replacement for modeling but a great complement. With better controlled environment I think scanning quality can be done in the sub millimeter accuracy. 4K video capture could also allow capture of movement or simply blendshape/texture animation.

For pro result you could need a well lit studio, high quality camera in the range of 41Mp ... even then it's not too outside what a indie can get, it's just more discipline

[Zaphos]

Cool!  Yeah that's probably an easier case than what I was trying to handle.

(I can't really tell how good the model is from that one image but since you sound happy with it I'll assume it's ok ...)

BTW if you're talking about scans of static objects, especially indoors, then to fairly compare to the Kinect you should really compare to something like KinectFusion: http://research.microsoft.com/en-us/projects/surfacerecon/ ( http://msdn.microsoft.com/en-us/library/dn188670.aspx ) 


A single depth image from kinect is generally pretty noisy but if you move the kinect around and fuse multiple depth images together, you can get a much higher-quality result.

[gimymblert]

Given I took the image loosely with no care at a low resolution and there was plenty noise in the environment (lot of specular light)and how I too the image (blur) so it's rather good. And that's with material that are difficult to capture (satin with anisotropic quality). Not sure it was the easier case : P not the worst too as I removed any transparent things that aren't great with this technique at all!

You can also use IR light to capture images and compose them, to avoid some light pollution, it's rather flexible.

Kinect Fusion use a LOT of images, it's a video feed The advantage is it's real time but not more precise!
Precision depend on the pixel coverage, and with the kinect they move it relatively close to objects (to compensate the low resolution of 320x240), I took all my images at a higher distance. It depend also on your camera range of focus. We have a kinect too but it's not as satisfying and quick.

With photogrammetry you control the quality with the device you choose so ...
The main drawbacks is that it is slower to compute for volume convergence.

That's why I seeking mobile like captor to build a scanner, they work well at short range and have infinite focus by default.


It seems that's you are an engineer, you can hijack the open source algo in C++ and build a kinect equivalent using the known pattern trick to speed up the things and using the IR domain with a video feed. That's something I can't do ... I'm no engineer I just try by piercing data from the net But this is over my own competence.

http://opensourcephotogrammetry.blogspot.com/2013/08/openmvg-c-open-source-multiple-view.html

If you do something I'm interested



BTW you can see pro application and result there:

An indie game use it to great result using the photoscan soft and one camera.
http://www.theastronauts.com/2014/03/visual-revolution-vanishing-ethan-carter/


Used in metal gear V to capture face notably the same soft that the indie above





Studio with multiple camera for instant capture or the brute force way of doing it.


The most advance and extreme version but basically the same things and basics
http://www.fxguide.com/featured/the-art-of-digital-faces-at-ict-from-digital-emily-to-digital-ira/

But even that is kinda accessible to clever people (see my link to material capture using a consumer camera).
It's a matter to alternate light and picture based on polarized light direction relative to the lens filter.


Basically I want to do a cheap version of the multiple camera setting with cheap captor, the parameter for quality are lower than for a pure artistic camera so we can cut cost.

[gimymblert]

So instead of taking a lot of photo, just take a camera feed with a good camera and then feed all the frame to a soft.

[Zaphos]

For a camera feed I think videotrace is interesting: http://punchcard.com.au/  (A lot of manual work though!)

Kinect Fusion use a LOT of images, it's a video feed The advantage is it's real time but not more precise!

Of course it's a lot of images   My point was that it shouldn't be that different, in terms of capture time, from moving your camera around and snapping 18 pictures -- which is why I said that should be the point of comparison to photogrammetry, rather than the animation capture you mentioned a while back.  Re being more precise, all I meant is that KinectFusion is more precise than 'raw' Kinect data: By fusing a bunch of low-resolution depth images into one model, you can get a higher resolution result.

[gimymblert]

There is also that from google



http://procrastineering.blogspot.com/


I knew about video trace, but that's I do with my acquired volume data on a regular 3D soft. Retopology is easier on an actual 3D model, the tools is great however.

I'll concede that kinect win on the sheer convenience side.

But the accuracy is not an issue for vanilla photogrammetry too, free soft like meshlab allow to merge and manipulate different volume or cloud point data. But by that point I think we can mixed the capture data with any of the techniques.

However the field is advancing really quickly, in five year is will be completely mature with great tools.

[Schoq]

https://twitter.com/congressedits

[gimymblert]

http://jemthemovie.com/

[Leon Fook]

you guy lose.

[Alevice]

I didnt realize that Egoraptor had released a Sequelitis comparing TLOZ, ALttP and OoT, and a bit of SS. Unfortunately the structure of the "show" gives no room to point out how the Souls games are basically modern oldschool TLOZ

[Leon Fook]

it's a not good comparison by him, although he got some point with the enemy. zelda is a time waster, regardless linear gameplay or not.

[ink.inc]

the points he makes are exactly why i get bored with (almost) every single zelda ever

[SirNiko]

How are souls games modern zelda?

The appeal of D*Souls comes from the passive storytelling (and appealing atmosphere), the unforgiving combat, the deeply integrated PvP and the high level of discovery in character customization, none of which have ever been a staple of the Zelda series even in the early incarnations. (Zelda 2 being an outlier for sending you back to Zelda's Palace after three deaths including insta-death pits, but even then you keep your XP (Unless you play the Japanese version).)

D*Souls builds a little off the combat of the 3D Zelda titles, but aside from lock-on and strafe, they have nothing in common. The things that draw a person to play each series seem pretty different to me.

[gimymblert]

well:
1. The story telling told by lore hint that give a much bigger and mysterious picture of the world (especially oot)
2. Non linear exploration with hidden secrets
3. Combat basics

The only difference is puzzle replace by customization and pvp, but DS is closer to zelda 1 and 2 that would have evolve if they kept their former philosophies.

that and zelda stop being revolutionary after WW (and WW was more evolutinary)