https://github.com/spandan-madan/gans_vs_humans
https://github.com/spandan-madan/gans_vs_humans
Last synced: 7 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/spandan-madan/gans_vs_humans
- Owner: Spandan-Madan
- Created: 2019-07-31T00:05:07.000Z (about 6 years ago)
- Default Branch: master
- Last Pushed: 2019-09-30T18:10:13.000Z (about 6 years ago)
- Last Synced: 2025-01-26T12:41:24.372Z (9 months ago)
- Language: Jupyter Notebook
- Size: 38.9 MB
- Stars: 1
- Watchers: 3
- Forks: 0
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# NNs_vs_Humans
Recently, GANs have shown great promise in generating fake faces, which look very much like real people. If you don't believe me, take a look at this website - thispersondoesnotexist.com
So, it is clear that GANs can fool humans (at least somewhat) when it comes to faces. However, sitting at a coffee shop on a sunday I was wondering if they can fool other Neural Networks like classification CNNs as well? So I decided to do this experiment to generate lots of faces using GANs and try to classify them using CNNs.
Turns out they can't fool a CNN. I thought this was pretty interesting, so I decided to publish my code base here for others.
Ideas :
- Masking facial features one by one (eyes, nose etc) and seeing which one breaks humans vs NN's more. Would likely require a pipeline for training on large number of datasets - as NN would need to be re-trained on new masked datasets.