Emotion Detection using deep unbaised CONV-net!!
The dataset used here is the famous FER2013 dataset
from kaggle's FER challenge of 2013
Zip File : from here 92MB
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
- Python
- Jupyter
- Keras
- Tensorflow
- Matplotlib
- Pandas
- Numpy
- tqdm
- Download the data and unzip it as
FER2013
dir. - Clone this repo.
- Add the full path of
fer2013.csv
intocell 3
ofFER2013-model1.ipynb
. - Run the file or use the pretrained model weights.
The Layers for the network :
- Change the path to weights folder in
cell3
.
Testset Accuracy :
Some Images prediction :
Happy
Emotion is the most detected, as it has most number of examplesSad
,Surprise
,Neutral
andAnger
are also good in detecting due to enough examples.Fear
andDisgust
perform worse, possible reasons : Less training examples and fordisgust
: pretty similar toanger
features.Sad
emotions are also closely detected asneutral
, cuz its hard to distinguish them with just this much data.