Crossing the uncanny valley WIP

#1

latest update :
[u][b]

[/b][/u]

[u][b]

               [/b][/u]So, this is gonna be sort of a making-of'ish, tutorial'ish thread on how to create an accurate face scan for anybody, basically all you need is a digital camera and internet connection. 
                             
                             So for me the plan was to create an one-to-one digital copy of a real life protagonist (from the neck up) for a DIY facial mocap later on. 
                             
                             To the scanning part, i came across a free service provided by autodesk called Project PhotoFly (im sure many of you heard of it and tried it). The quality of output this service provides is imho mindblowing.
                          
                          The first crucial step is provide photofly with enough data 
                     
                             [img]http://img28.imageshack.us/img28/3589/fotosqc.jpg[/img]
                           I shot about 90 to 100 18 megapixel photos from every possible angle. 
                           They were shot with 50mm portrait lens. I aimed for shots as sharp as possible with minimal depth of field (f/8 ; 1/250s). I also found myself getting much better results with more closeup shots (on mouth area,  eyes area, nose area etc) in stead of trying to fit the whole face into the image like most people do. The difference in quality was really significant
                          
                          This is how the model look in photofly viewport:
                         
                          [img]http://img89.imageshack.us/img89/9349/phtfl.jpg[/img]
                          This image is only for illustration purposes. The golder rule is 'less is more' so i didnt use all my taken photographs, i just used around 50 photos, its a kinda trial & error process, choosing and uploading different photographs and checking the model if the geometry is glitch free and clean. And if its not , then figure out which photo is causing problems and remove it, or manually restitch it
                         
                         This is the final scan i end up working with:
                         
                         [img]http://img856.imageshack.us/img856/714/modelw.jpg[/img]
                         Not bad for a free and easy solution if you ask me :) not bad at all. The texture is ok too
                     The eyebrows and eye area will always cause problems because of eyelashes and facial hair, but thats absolutely fine. 
                     
                     [img]http://img191.imageshack.us/img191/9141/zbrush.jpg[/img]
                     Some minor retouching in zbrush with the trim dynamic brush and smooth subdivision, which works like magic, smoothing the surface without destroying nice details 
                     
                     And of we go to topogun for retopology...
#2


Before starting the actual retopo process, i marked out some key features i wanted my mesh to follow. Besides the obvious ones, like mouth and eye edgeloops, i also wanted my mesh to follow the jaw line, the nose-labial fold, the nose geometry collapsing on the bridge of the nose so the forehead geometry would be clean and smooth

[img]http://img843.imageshack.us/img843/6383/topogun.jpg[/img]

The final topology looks something like this. i tried to make the polygons as even as possible. i must say im pretty happy with the end result. The trouble areas like the mouth nose and mainly eyes will be modeled in maya by hand


The exported model looks very clean, smooth and organic, the anatomy underneath the skin is well defined, and thats everything i was hoping for

#3

Another awesome feature of photofly is that you can export not only the generated mesh, but also the whole scene with every used camere, so you can then model the face from many many angles and lock it dead. Much more accurate than using orto views i guess :slight_smile:


Imho the retopologized mesh is about >95% accurate with the actual real life protagonist


All you need to do is chose the desired camera and import the imageplane accordingly. and trust me, youll have plenty of cameras to chose from

Hint : another neat way of preparing a scene for modeling is by using PFTrack,but thats a different story


#4

I also did a small side quest, and exported normal, displace, occlusion maps on a uv’d model
then i combined the extracted geometry-normal map with texture-normal map (using crazybump) for high fidelity details. the results again look pretty neat:

Next up : good’ol fashion modeling

#5

This looks fantastic! Can’t wait to see more of it!

#6

Incredible! This is pretty cool and ever more so, because it’s free! Keep them coming…

#7


this is how a crazy person prepares a scene for modeling. ,)

#8

building the eye geometry.

mobile workstation + big external monitor = my best modeling experience ever

#9

I’m following this thread with huge interest. It seems that the tool is really working great and the model already has a big resemblance. Makes me want to check it out myself. Keep us updated…

#10

the left eye geometry is finished. im pretty happy with it so far.
only minor adjustments left to do

fits like a glove ,)

+

#11

Wow, this makes a really nice looking normal map. This is really interesting. Thanks for sharing

#12

I’ve just finished the geometry for both of the eyes.
Next up, wrap up the nose and mouth to complete the face

+

#13

i finally had some spare time, and so i did some minor modelling to finish the nostrils and mouth and adjust a thing or two

Its just a detail but it brings out the character so much. after doing this, i can definitely feel the very accurate likeness in the face…which is kinda creepy since the guy is sitting right next to me. :scream:

and like before, the model still sits very very nicely

some big quests before christmas ahead. but after that i cant wait to start with the ear ‘scanning’ and modeling.

  • C&C are of course welcome
#14

I don’t know how realistic you’re going, or how close up you’re going to render, but I might suggest building the tear duct as a separate piece of geometry. It’ll handle your shaders and blinking better. But only if you will ever be up close enough to notice it.

#15

@TechnicallyArtisti

now when im thinking about it… i just NEED to do so. the shaders are one reason, and also during animation and eye rigging it wont be getting in the way.

great suggestion thanks a lot

#16

today i did some quick tests enhancing the scan data by extracting the fine skin details from the textures.

#17

moving on

#18

Wow this technique looks amazing.
When i get some time i will most certainly try it out myself.

I’m gonna keep watching this thread.

#19

i made an ear ‘scan’ today.

the main thing i was interested into were the calibrated cameras.
the ear itself will be modeled by hand, the scanned mesh serves only as a guide and rough reference

#20

This thread is indeed very inspiring. Definitely makes me want to try that out. I heard about an other software like photofly which is called ScannerKiller. I have never tried one of those tools but i wonder if they produce similar quality. Did you try multiple tools?
I’m curious about the next updates :slight_smile: