Menu

Image recognition by DNN and CNN

Simple Deep neural networks and Convolutional Neural Networks used to recognise 2D images of deformed objects generated with Houdini.

Details

Few different representations of objects ware gathered to determinate which will work the best:

  • Voxels

voxel

  • Projection on plane (distance from plane surface)

plane

  • Projection on cylinder

cylinder

  • Projection on sphere

sphere

net

Visualisation of filters

To understand networks a little bit better I created a visualization of filters and activations:

filters

It’s in /obj/visualize_filters of recreate_pivot_volume.hiplc

Testing

Test is in /obj/test of recreate_pivot_volume.hiplc

test

Colored lines on the right part of this screenshot represent orientations predicted by different networks.