https://github.com/sintef/pfv

Principal Feature Visualization for convolutional neural networks
https://github.com/sintef/pfv

Last synced: about 2 months ago
JSON representation

Principal Feature Visualization for convolutional neural networks

• Host: GitHub
• URL: https://github.com/sintef/pfv
• Owner: SINTEF
• Created: 2020-07-16T07:16:22.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2021-01-28T16:44:03.000Z (almost 4 years ago)
• Last Synced: 2023-10-20T23:08:48.096Z (about 1 year ago)
• Language: Python
• Homepage:
• Size: 2.02 MB
• Stars: 11
• Watchers: 2
• Forks: 5
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# 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:



## The paper

This method was presented at ECCV 2020. Please see [the full paper](https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123680018.pdf) and [supplementary material](http://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123680018-supp.pdf) for more information about our method.

If you find this useful, please cite:

```

@inproceedings{bakken2020principal,

  title={Principal Feature Visualisation in Convolutional Neural Networks},

  author={Bakken, Marianne and Kvam, Johannes and Stepanov, Alexey A and Berge, Asbj{\o}rn},

  booktitle={European Conference on Computer Vision},

  pages={18--31},

  year={2020},

  organization={Springer}

}

```