sridharsola/fer

Facial Expression Recognition using ResNet-18 in PyTorch. Trained and tested on RAFDB and FERPlus

FER

This project uses deep learning to perfrom facial expression recognition. We used ResNet18 pretrained on MSCeleb-1M as our network. Our experiments used two benchmark datasets: RAFDB and FERPlus.
Further, a face mask was synthetically placed on images of these datasets so we could run experiments on masked images.
The results of these experiments served as baseline results for further studies (see projects on Knowledge Distillation and Contrastive Learning).

Datasets

RAFDB Sample
FERPlus Sample

Experiments

Experiment 1 – Training ResNet18 on non-masked images.
Experiment 2 – Training ResNet18 on masked images.
Experiment 3 – Training ResNet18 on masked images when pre-trained on non-masked images.
Experiment 4 – Training ResNet18 on Mixed (masked and non-masked) dataset.

Results