https://github.com/antonior92/aliasing-in-cnns

How convolutional neural networks deal with aliasing
https://github.com/antonior92/aliasing-in-cnns

aliasing convolutional-neural-networks deep-learning deep-neural-networks signal-processing

Last synced: about 1 month ago
JSON representation

How convolutional neural networks deal with aliasing

• Host: GitHub
• URL: https://github.com/antonior92/aliasing-in-cnns
• Owner: antonior92
• License: mit
• Created: 2020-10-19T21:20:34.000Z (over 4 years ago)
• Default Branch: main
• Last Pushed: 2021-02-16T14:12:06.000Z (almost 4 years ago)
• Last Synced: 2024-10-31T01:13:39.885Z (3 months ago)
• Topics: aliasing, convolutional-neural-networks, deep-learning, deep-neural-networks, signal-processing
• Language: Python
• Homepage:
• Size: 281 KB
• Stars: 4
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# How Convolutional Neural Networks Deal with Aliasing

Python scripts for reproducing the results from the paper: ["How Convolutional Neural Networks Deal with Aliasing"](https://arxiv.org/abs/2102.07757).

```

Antônio H. Ribeiro and Thomas B. Schön "How Convolutional Neural Networks Deal with Aliasing". IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021.

```

```

@inproceedings{ribeiro_how_2021,

author={Ant\^onio H. Ribeiro and Thomas B. Sch\"on},

title={How Convolutional Neural Networks Deal with

Aliasing},

year={2021},

publisher={IEEE},

booktitle={2021 {IEEE} International Conference on Acoustics, Speech and Signal Processing, {ICASSP}}

```

Preprint: https://arxiv.org/abs/2102.07757

------

The folders in this repository contain two experiments:

1. [classifying-oscillations](./classifying-oscillations): toy example designed to assess the ability

    of convolutional neural networks to resolve between different frequencies at its input.

1. [quantifying-aliasing](./quantifying-aliasing): Scripts for quantifying to what extent aliasing takes 

    place in the intermediate layers of the neural network

    

Requirements

-----------

The file `requirements.txt` contains the python modules required. 

The versions specified are the ones the code has been tested on. Nonetheless,

I believe lower versions of most packages should also work. One exception is matplotlib where I observed that using versions different than

the 3.2.1 might yield minor changes (namely, different axis ticks).

Finally, some experiments also require ImageNet validation set. I include basic

instructions for applying for the license, downloading the dataset and extracting it [here](./quantifying-aliasing/README.md).