/PFV

Principal Feature Visualization for convolutional neural networks

Primary LanguagePython

Principal Feature Visualization (PFV)

Principal feature visualization is a visualization technique for convolutional neural networks that highlights the contrasting features in a batch of images. It produces one RGB heatmap per input image.

Dependencies

  • pytorch
  • numpy

Additional dependencies for the demo:

  • torchvision
  • matplotlib
  • pillow

Getting started

Install the dependencies listed above, and run the example in demo.py: python demo.py

Example

A trained network shows good localization:

But an untrained (re-initialized) network shows scrambled output, as expected: