SimCLR on MNIST CIFAR10 STL10

This project is a SimCLR training pipeline compatible with :

• MNIST Dataset
• CIFAR10 Dataset
• TMarks Dataset
• Shearo Dataset

To use this repo, install library / dependancies :

conda create --name vae
conda activate vae
conda install pytorch torchvision cudatoolkit=11.3 -c pytorch
pip install -r requirements.txt

To launch a training:

python train_CLR.py    # train contrastive learning

In order to modify the dataset, you can use use argparse parameters.

SimCLR

In this repo, we have implemented :

What	Code
A Convolutive Encoder	$f(.)$
A Non Linear Projection Head	$g(.)$
Augmentations	$(t_1(.),t_2(.))$

Let's say we have an image $X$.
$X_1 = t_1(X)$ and $X_2 = t_2(X)$

We have features space :
$h_1=f(X_1)$ and $h_2=f(X_2)$

And the prediction space :
$z_1=g(h_1)$ and $z_2=g(h_2)$