You are on page 1of 5

Mid-Term Project

Emotion Recognition using Deep


Learning
2

The Dataset(1)
• Kaggle FER2013 Dataset
• Facial Expression Recognition Challenge
• 7 classes: (0=Angry, 1=Disgust, 2=Fear, 3=Happy, 4=Sad, 5=Surprise,
6=Neutral)
3

The Dataset(2)
• Train & Testing Data
• https://www.kaggle.com/c/challenges-in-representation-learning-facial-ex
pression-recognition-challenge/data
• https://www.kaggle.com/datasets/dntai1983/fer2013-onehot
• World-best accuracy: 69%(public), 71%(private)
4

CNN Architecture
• Input/Output & Environment
• Input: 48x48 image
• Output: 7D vector  emotion classes
• Performance Improvements
• CNN model design(Conv, MaxPooling, Dropuout, FC, …)
• Training techniques: optimization algorithm, data augmentation, …
• Activation function: ReLU, sigmoid, …
Pay Attention
• Refer to the sample code
• Submission
• Report
• Brief summary for source codes
• Results for many experiments with various parameter settings
• Visualize the final convolution filters
• Analysis and explanation about the results in your own way
• Python Source Code using Keras

You might also like