Emotion-detection

Emotion Detection using deep unbaised CONV-net!!

Website : https://ash1998.github.io/Emotion-detection/

Getting Data

The dataset used here is the famous FER2013 dataset from kaggle’s FER challenge of 2013

Download

Zip File : from here 92MB

Using API :

1.Install Kaggle from github
2.Use the command in terminal kaggle competitions download -c challenges-in-representation-learning-facial-expression-recognition-challenge

Docs on Kaggle API usage : github | kaggle

Dependencies

1. Python
2. Jupyter
3. Keras
4. Tensorflow
5. Matplotlib
6. Pandas
7. Numpy
8. tqdm

Usage

For model1:

1. Download the data and unzip it as FER2013 dir.
2. Clone this repo.
3. Add the full path of fer2013.csv into cell 3 of FER2013-model1.ipynb.
4. Run the file or use the pretrained model weights.

   Using pre-trained weights:

   The Layers for the network :

For EDA notebook :

1. Change the path to weights folder in cell3.

   Images

   Testset Accuracy :

Some Images prediction :

EDA confusion Matrix and Scores:

ED Analysis:

1. Happy Emotion is the most detected, as it has most number of examples
2. Sad, Surprise, Neutral and Anger are also good in detecting due to enough examples.
3. Fear and Disgust perform worse, possible reasons : Less training examples and for disgust: pretty similar to anger features.
4. Sad emotions are also closely detected as neutral, cuz its hard to distinguish them with just this much data.