cl626 / FER_Conv_res

facial emotion recongnition with cnn and resnet

pytorch-facial-expression-recognition

Lightweight facial emotion recognition.

• Pytorch implementation for Mini Xception inspired by Keras implementation.
• Model size is about 250KB

Trained Model

Trained by FER2013 dataset.

• Private Data : 66%
• Public Data : 64%

Here is the result of sample image.
Emotion | Probability | Guided Backprop | Grad-Cam | Guided Grad-Cam

Retrain

1. see here to prepare dataset.

2. execute train.py

cd src
python train.py
python check.py  #check.py supports cpu only

Reference

• Grad-CAM
• Data Augmentation / Optimizer