The aim is to build and test CNNs to tackle the FER (Facial Emotion Recognition) problem.
The project expects the fer2013 dataset to be present as a csv file named fer2013.csv under ./data/ directory.
- Download the dataset from here
- Untar the file and place the fer2013.csv fil under the ./data/ directory
The project makes use of Python 2.7 and PyTorch.
PyTorch can installed from here
Other dependencies can be found in the requirements.txt file
Run the following command in the project home to install the python libraries.
pip install -r ./requirements.txt
The network can be trained by running the train.py in the home directory of the project.
The architecture of the network can be specified through the --model parameter. An example is shown below.
python train.py --model resnet20
Valid models include bkvgg12, resnet20 and cnn_sift
The list of all arguments that can be provided to train.pycan be see by running the following command
python train.py --help
There are arguments that can be used to specify the epcohs, batch_size, optmizer etc.
- tensorboard_logger (option)
- TensorFlow (option)