ImageReflect
Image3DSlices
SetAlphaChannel
NetModel
NetEncoder
NetGraph

Machine Learning
Image Processing
Computer Vision

3D Facial Reconstruction from 2D Images

Use neural networks to estimate the shape of a face with unguided volumetric regression

Open in Cloud
Download Notebook

Start with an image of a face:

In[1]:=

face=

;

Get the following unguided volumetric regression neural net from the Wolfram Neural Net Repository:

In[2]:=

net=NetModel["Unguided Volumetric Regression Net for 3D Face Reconstruction"]

Out[2]=

NetChain



Input port:	image
Output port:	array (size: 200×192×192)



Evaluate the network on the face image. The network output is converted into a 3D image:

In[3]:=

face3D=Image3D[net[face]]

Out[3]=

Resize the 3D volume to have the same width and height as the 2D image:

In[4]:=

reconstruction=Blur@ImageReflectImageResize[face3D,Append[ImageDimensions[face],105]],

...



Out[4]=

Multiply the original image with each slice to assign the proper gray level:

In[5]:=

coloredSlices=Map[SetAlphaChannel[face,#]&,Image3DSlices[reconstruction,All,2]];

Display some of the slices:

In[6]:=

Part[coloredSlices,{85,75,65,55}]

Out[6]=





Reconstruct the 3D image from the list of slices:

In[7]:=

ImageReflectImage3DcoloredSlices,

,

...



Out[7]=

Related Symbols

ImageReflect
Image3DSlices
SetAlphaChannel
NetModel
NetEncoder
NetGraph

Publisher Information

Contributed by: Wolfram Staff

Wolfram Language Example Repository

3D Facial Reconstruction from 2D Images

See Also

Publisher Information