https://github.com/agdhruv/reconstructing-obfuscations
A machine learning approach to beating facial image obfuscation
https://github.com/agdhruv/reconstructing-obfuscations
Last synced: 3 months ago
JSON representation
A machine learning approach to beating facial image obfuscation
- Host: GitHub
- URL: https://github.com/agdhruv/reconstructing-obfuscations
- Owner: agdhruv
- Created: 2019-05-27T16:17:29.000Z (about 6 years ago)
- Default Branch: master
- Last Pushed: 2020-01-08T15:20:19.000Z (over 5 years ago)
- Last Synced: 2025-01-25T22:13:05.993Z (4 months ago)
- Language: Jupyter Notebook
- Size: 55.9 MB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# reconstructing-obfuscations
A machine learning approach to beating facial image obfuscation (pixelation, blur, block) using Autoencoders. Explanations for our wok can be found in `Report.pdf`. This was our mid-semester project for Advanced Machine Learning (CS-408) at Ashoka University (in Spring 2019).
Trained models are available in the `trained_models/` directory. You can use them in conjunction with the `demos/` to see our models in action. If required, read below to replicate the training process. Folder names are self-explanatory.
Authors: [Prakhar Jain](https://github.com/prakhar171), [Harsh Karamchandani](https://github.com/harshk0525), [Dhruv Agarwal](https://github.com/agdhruv)
## Dataset
We use the Labelled Faces in the Wild (LFW) dataset from [here](http://vis-www.cs.umass.edu/lfw/#download). Download the dataset and place it in the root folder of the project in a folder named `lfw/`.
## How to Run
Note: This specifies the order of execution to replicate the project. File specific instructions are in the notebooks themselves.
Training Process:
1. Download the dataset and place it in `generate_datasets/` in a folder named `lfw/`. Currently, there is a partial LFW dataset present there.
2. Generate training datasets in pickle form.
- Run `generate_datasets/dataset.ipynb` after making necessary modifications.
- This will generate pickle files containing training images in `training_datasets/` in the root of the project.
3. Train (and test) the model(s).
- Run `train.ipynb` in the root of the project after making necessary modifications.
- You will need to install Tensorflow, Scikit-learn, Matplotlib, etc. to run this notebook.
Demo Process (demoing trained models):
1. First generate pickle files containing the files you want demo the models on. These files must be placed inside `generate_datasets/demo_images/`. You will see a few sample images (these are pictures of the authors).
- Run `generate_datasets/dataset-for-demo.ipynb` after making necessary changes.
- This will generate pickle files containing demo images in `demo_datasets/` in the root of the project. You will see sample pickle files in this folder, *which you can use to demo the project without going through the training process*.
2. Run the demos inside `demos/` in the project root.
- Run the three demo files after making necessary modifications.