ResEmoteNet: Bridging Accuracy and Loss Reduction in Facial Emotion Recognition

A new network that helps in extracting facial features and predict the emotion labels.

The emotion labels in this project are:

• Happiness 😀
• Surprise 😦
• Anger 😠
• Sadness ☹️
• Disgust 🤢
• Fear 😨
• Neutral 😐

Table of Content:

• Installation
• Usage
• Checkpoints
• Results
• License

Installation

1. Create a Conda environment.

conda create --n "fer"
conda activate fer

2. Install Python v3.8 using Conda.

conda install python=3.8

3. Clone the repository.

git clone https://github.com/ArnabKumarRoy02/ResEmoteNet.git

4. Install the required libraries.

pip install -r requirement.txt

Usage

Run the file.

cd train_files
python ResEmoteNet_train.py

Checkpoints

All of the checkpoint models for FER2013, RAF-DB and AffectNet-7 can be found here.

Results

• FER2013:
  • Testing Accuracy: 79.79% (SoTA - 76.82%)
• CK+:
  • Testing Accuracy: 100% (SoTA - 100%)
• RAF-DB:
  • Testing Accuracy: 94.76% (SoTA - 92.57%)
• FERPlus:
  • Testing Accuracy: 91.64% (SoTA - 95.55%)
• AffectNet (7 emotions):
  • Testing Accuracy: 72.93% (SoTA - 69.4%)
• ExpW:
  • Testing Accuracy: 75.67%

License

This repository is licensed under the MIT License. See the LICENSE file for more details.

Cite our paper:

@ARTICLE{10812829,
  author={Roy, Arnab Kumar and Kathania, Hemant Kumar and Sharma, Adhitiya and Dey, Abhishek and Ansari, Md. Sarfaraj Alam},
  journal={IEEE Signal Processing Letters}, 
  title={ResEmoteNet: Bridging Accuracy and Loss Reduction in Facial Emotion Recognition}, 
  year={2024},
  pages={1-5},
  keywords={Emotion recognition;Feature extraction;Convolutional neural networks;Accuracy;Training;Computer architecture;Residual neural networks;Facial features;Face recognition;Facial Emotion Recognition;Convolutional Neural Network;Squeeze and Excitation Network;Residual Network},
  doi={10.1109/LSP.2024.3521321}
}