Jamal, Aidan, Chidansh, and Ahmed
# | Paper Title | Features(s) | Implementation |
---|---|---|---|
1 | Toward Fairness in Face Matching Algorithms | Gender | Adversarial deep learning |
# | Title | Contributor(s) | Description |
---|---|---|---|
1 | main.py | Jamal | Training |
2 | main.ipynb | Jamal, Aidan | Annotated main.py |
3 | low_high_celeb_adversal_test_all.py | Jamal | CelebA Testing |
4 | test_main.py | Jamal | UMD Faces Testing |
Name | Description |
---|---|
CelebA | 200K celebrity images with 40+ attributes. |
UMD Faces | University of Maryland images with gender and box feature labels. |
Additional data sets can be found on Kaggle
- Joint Feature and Similarity Deep Learning for Vehicle Re-identification
- Variational Fair Autoencoder
- A Cloud-guided Feature Extraction Approach for Image Retrieval in Mobile Edge Computing
- HyperFace: A Deep Multi-task Learning Framework for Face Detection, Landmark Localization, Pose Estimation, and Gender Recognition
- Deep Feature Consistent Variational Autoencoder
- Bala's Intro to Amarel Video
- Kaltura Videos
- OnDemand
- OARC
- User Guide
- Add the following to SLURM to use GPUs:
#SBATCH --gres=gpus:1
#SBATCH -partition=gpu
- To clone the repo, enter below snippet into your command line
git clone https://github.com/Behavioral-Informatics-Lab/Face_Matching.git
- Remember to pull to avoid version conflicts!
git pull
- To push your changes to the git repo, enter the following (or a similar variant) in your command line:
git add .
git commit -m "Your comment"
git push
- If pushes are taking too long, then consider
git config http.postBuffer 524288000
- Locally install the images ONLY from the CelebA data set. This should be a zip file. The annotations should already be a part of the repo. The file structure:
├── Checkpoints
└── Data Sets
└── CelebA
├── annot
└── img_align_celeba
└── [INSERT IMAGES HERE]
└── UMD Faces
- first use (0), after checkpoint saved (1)
load_net = 0
- CPU (0), GPU (1)
use_cuda = 0
- Modify train, validation, and test ratios accordingly
train_ratio = 0
val_ratio = 0
test_ratio = 1-train_ratio-val_ratio
- Make sure you are using the correct version of pytorch for running Jamal's code
pip install torch===1.6.0 torchvision===0.7.0 -f https://download.pytorch.org/whl/torch_stable.html
- Run code (here is what my command line looks like)
PS C:\Users\austr\Desktop\BIL Face Matching Alg Fairness\Face_Matching> & C:/Users/austr/anaconda3/python.exe "c:/Users/austr/Desktop/BIL Face Matching Alg Fairness/Face_Matching/Paper1Scripts/low_high_celeb_adversal_test_all.py"