Recognition of 3D object is common task for human but hard for computer.
Interesting report was published by researchers work in Princeton University.
URL is following.
They developed 3D shape prediction system named ‘3D ShapeNets’.
I was interested the system because it collect image data from one side and then, the system predict next view point to minimise uncertain data.
The author described ….
For example, humans do not need to see the legs of a table to know that they are there and potentially what they might look like behind the visible surface
To do that, next best view prediction is key.
They tried to predict 3D object like furniture i.e. table, sofa, bed, etc….
And 3D-Shapenets (Based powerful CNN) showed good performance.
In this report, I could know descriptor named Light Field Descriptor(LFD), Spherical Harmonic Descriptor(SHD), and voxel ( like pixel in 2D ). 😉
In drug discovery project, 3D shape of molecule was predicted from conformational search. So, It’s different in this case.
But 3D shape recognition is interesting and deep area.
Also I found amazing report in J Chem inf models.
The system uses kinect and source code was uploaded github.
Think molecule as 3D!