Facial-Expression-Recognition

A Machine Learning system that can detect facial emotions on human faces.

Description

Implement a CNN model to detect facial emotions on human faces.

Getting Started

Dependecies

• Install all project dependecies using pip command

pip3 install -r requirements.txt

Dataset

• Step 1: Sign up for Kaggle.
• Step 2: Get Kaggle API credentials.
  • Go to "Account Setting".
  • Scroll down to the section labeled "API" and click on the "Create New API Token" button.
  • This will download a file named "kaggle.json" containing your API credentials.
  • Keep "kaggle.json" in "~/.kaggle/" to be able to use Kaggle API.
• Step 3: Download the dataset using Kaggle API.

kaggle datasets download -d msambare/fer2013

• Step 4: Unzip the dataset.

unzip fer2013.zip

• Step 5: Combine the "test" and "train" folders into a folder called "datasets"

Dependecies

• Install all project dependecies using pip command

pip3 install -r requirements.txt

Train the models

• Step 1: Choose the device to run the model
  • Type nvidia-smi to see all available GPUs in the CS machine.
  • Choose the one that is not being used by other processes.
• Step 2: Run the following command to train the model. Notice we only support one of these four models: VGG11, VGG13, VGG16 and VGG19.

CUDA_VISIBLE_DEVICES = <gpu_number> python3 train.py --model <model_name> --num-epochs <number of epochs>

• Step 3: After the training, the model will save its checkpoint inside the models folder and the accuracy vs loss plot under the name acc_and_loss_<model_name>.png

Authors

Contributors names and contact info

• Bao Nguyen (btn6364@rit.edu)
• Neha Kulkarni (nk4349@rit.edu)