[QUOTE]Machine Learning research from University of Nottingham School of Computer Science can generate a 3D model of a human face from an image using neural networks:
3D face reconstruction is a fundamental Computer Vision problem of extraordinary difficulty. Current systems often assume the availability of multiple facial images (sometimes from the same subject) as input, and must address a number of methodological challenges such as establishing dense correspondences across large facial poses, expressions, and non-uniform illumination. In general these methods require complex and inefficient pipelines for model building and fitting. In this work, we propose to address many of these limitations by training a Convolutional Neural Network (CNN) on an appropriate dataset consisting of 2D images and 3D facial models or scans. Our CNN works with just a single 2D facial image, does not require accurate alignment nor establishes dense correspondence between images, works for arbitrary facial poses and expressions, and can be used to reconstruct the whole 3D facial geometry (including the non-visible parts of the face) bypassing the construction (during training) and fitting (during testing) of a 3D Morphable Model. We achieve this via a simple CNN architecture that performs direct regression of a volumetric representation of the 3D facial geometry from a single 2D image. We also demonstrate how the related task of facial landmark localization can be incorporated into the proposed framework and help improve reconstruction quality, especially for the cases of large poses and facial expressions. [/QUOTE]
Basically, the system takes a frontal photograph of a person and tries to construct a 3d model of it. Some examples:
[IMG]https://68.media.tumblr.com/1df63f82c754a2723bcc65eb5528c852/tumblr_owcc3mInAo1qav3uso1_540.gif[/IMG]
[IMG]https://68.media.tumblr.com/445b57c2f2dedb57ea5e4957836a6bed/tumblr_owcc3mInAo1qav3uso3_r1_540.gif[/IMG]
[IMG]https://68.media.tumblr.com/0a36b0136b03b2a91018b396b51f53dc/tumblr_owcc3mInAo1qav3uso2_r1_540.gif[/IMG]
[vid]http://www.cs.nott.ac.uk/~psxasj/300vw.mp4[/vid]
[URL="http://aaronsplace.co.uk/papers/jackson2017recon/"]Here's the page about it[/URL]
and the [URL="http://www.cs.nott.ac.uk/~psxasj/3dme/"]demo itself[/URL]
Imagine how powerful this would be using 3 images to try and make it as accurate as possible
out of curiosity i wanted to see if it would recognize a half-life 2 model, since they were made using photos of faces
[IMG]https://my.mixtape.moe/bblfpb.png[/IMG]
tbh idk what i was expecting
[t]https://i.imgur.com/D0b3Bl5.png[/t]
[url]http://www.cs.nott.ac.uk/~psxasj/3dme/view.php?name=59bc902870a00[/url]
Bill Murray as Trump
[t]https://i.imgur.com/Uu5baBA.png[/t]
to
[t]https://i.imgur.com/hL3oziC.png[/t][t]https://i.imgur.com/OP6U9cX.png[/t]
[url]http://www.cs.nott.ac.uk/~psxasj/3dme/view.php?name=59bc90f940111[/url]
holy fuck
also
[url]http://www.cs.nott.ac.uk/~psxasj/3dme/view.php?name=59bc90863d393[/url]
[T]https://gyazo.com/58250c1f858567a7dd363a06a3eebec3.png[/T][T]https://gyazo.com/7bb5ca38fad256343ae06e62cac15f3f.png[/T]
[editline].[/editline]
Doesn't seem to handle bright white areas too well, making them transparent
[T]https://gyazo.com/e08e28caee142ac85a08e508cd2adf49.png[/T]
This is pretty cool. It doesn't handle glasses well but I'm not particularly surprised by that.
well can't force a false positive with my avatar :/
[img]https://my.mixtape.moe/nafkay.png[/img]
[editline]15th September 2017[/editline]
[img]https://my.mixtape.moe/wghyfq.png[/img]
[QUOTE=J!NX;52685944]Imagine how powerful this would be using 3 images to try and make it as accurate as possible[/QUOTE]
Photogrammetry is already a thing, but combining this with it would be incredible.
[QUOTE=ZombieDawgs;52686001]Photogrammetry is already a thing, but combining this with it would be incredible.[/QUOTE]
Yeah, usually though the issue is that it needs a crazy number of photos at 50 billion different angles just to work
This would effectively increase efficiency with far less photos
[IMG]https://i.imgur.com/n71yHL8.png[/IMG]
[IMG]https://i.imgur.com/qcPByrR.png[/IMG]
[IMG]https://i.imgur.com/8eO8tT2.png[/IMG]
[IMG]https://i.imgur.com/RZKkXdd.png[/IMG]
[IMG]https://i.imgur.com/fo3GGZl.png[/IMG]
:v: I should make this into a playermodel for GMod
has anyone able to download the .obj file? I haven't been able to.
[b]EDIT:[/b]
database might be overloaded with people using it, I'm able to download obj files right now.
[IMG]https://i.imgur.com/678trLw.png[/IMG]
[IMG]https://i.imgur.com/RZZ1iit.png[/IMG]
This is great.
[QUOTE=Egevened;52686030][IMG]https://i.imgur.com/678trLw.png[/IMG][/QUOTE]
[img]https://facepunch.com/image.php?u=77321&dateline=1468895977[/img]
best what I can do with the vodka Mike is holding.
[url]http://www.cs.nott.ac.uk/~psxasj/3dme/view.php?name=59bc9bdd28797[/url]
[t]https://i.imgur.com/Uhp8gjc.png[/t][t]https://i.ytimg.com/vi/AEe8NBOj2LI/hqdefault.jpg[/t]
the computer was fooled
[QUOTE=Durrsly;52685950]out of curiosity i wanted to see if it would recognize a half-life 2 model, since they were made using photos of faces
[IMG]https://my.mixtape.moe/bblfpb.png[/IMG]
tbh idk what i was expecting[/QUOTE]
[img]https://i.gyazo.com/309f6fe0b306d9b26cb3a3547214ac4f.png[/img]
Of course I had to test it with the face model now. :v:
[QUOTE=J!NX;52686007]Yeah, usually though the issue is that it needs a crazy number of photos at 50 billion different angles just to work
This would effectively increase efficiency with far less photos[/QUOTE]
well no to what can be defined the usual, there is a subclass of 3d reconstruction through just monocular vision, eg using just an images or videos.
What this shows is robustness.
here's something similar from different people.
[media]https://www.youtube.com/watch?v=u9UUWqVquXo[/media]
not to say you're wrong to this possibly being applied to other methods to increase efficiency with far less photos.
[QUOTE=exhale77;52686182]give her the 3-d[/QUOTE]
greatly underrated post
whitewashed kim jong un
[t]https://i.imgur.com/6Q3qLhQ.png[/t]
[t]https://i.imgur.com/jPkzbKn.png[/t]
blackface putin
[t]https://i.imgur.com/Sryhxr9.png[/t]
[t]https://i.imgur.com/IdtJes5.png[/t]
I know a good rye when I see one(this one came out really nice)
[t]https://i.imgur.com/BnUfyYy.png[/t]
we wuz friends
[t]https://i.imgur.com/WKeLPzT.png[/t]
