LetheSec / Fer2013-Facial-Emotion-Recognition-Pytorch

This method achieves SOTA single model accuracy of 73.70 % on FER2013 without using extra training data.
MIT License
84 stars 16 forks source link

Question about other model #9

Closed VietPT3502 closed 1 year ago

VietPT3502 commented 1 year ago

Hi, i want to ask if u have try bigger resnet like resnet50? will this help to increase performance or it will only leads to overfitting ?

LetheSec commented 1 year ago

I tried several model architectures below, but not including ResNet50.

(Due to the requirements of the course project, the results in the table below are not trained with the full original data set and, for reference only)

image