face-expression-recognition

A Facial Expression Recognition Model trained on multiple dataset. So far:

• CKPlus dataset Link
• Emotic dataet Link

Setup

create a conda environment by:

conda create -n faceexpr python=3.10
conda activate facexpr

and then instal requirements:

pip install -r requirements.txt

Training

1. For the CKPlus dataset

• Create dataset folder:

mkdir dataset

• Unzip the CK+.zip dataset and put the folder CK+ in the dataset folder
• Modify the config file ckplus.yml, placed in the ./configs/ directory, if needed
• training:

python train_ck.py

2. For the Emotic dataset

• Create dataset folder:

mkdir dataset

• Unzip the emotic.zip in the dataset folder, you will have a folder named Emotic, in which there will be emotic folder that has 4 folders inside
• Unzip the Annotations.zip in the Emotic folder, you will have a folder named Annotations
• You will have the following structure then:

├── ...
│   ├── emotic
│   |    ├── ade20k
│   |    ├── emodb_small
│   |    ├── framesdb
│   |    ├── mscoco 
│   ├── Annotations
│   |    ├── Annotations.mat

• To convert annotations from mat object to csv files and preprocess the data:

python ./codes/mat2py.py --data_dir ./dataset/Emotic/

See this repo for more info

• Modify the config file emotic.yml, placed in the ./configs/ directory, if needed
• training:

python train_emotic.py

All the logs and weights will be stored in logs/ folder (look at the train scripts args for more detail)

Inference

to be added soon ...

Note For Me:

The motion_encoder.py in the codes folder contains the neural network. It should be modified and is not compatible yet to all datasets