soi-hardly-human-2022

This project inteds at classifying face images based on emotions. There are 7 categories namely

1. Happy
2. Sad
3. Angry
4. Disgust
5. Neutral
6. Surprise
7. Fear

We have used a CNN which trains on fer2013 dataset. Link to the dataset is here.

Architecture of our CNN is described in SOI_HH_Docs.pdf file.

Instructions

1. Clone the repository using the following command.

git clone https://github.com/shashankp28/soi-hardly-human-2022.git

2. Next install the required libraries.

pip install -r requirements.txt

3. Download all 3 models from google drive using this link.
4. Copy these models into the model folder.

Please use a device with a camera to proceed.

File descriptions

1. SOI_HH_Augmentaion:
   You need to have the dataset with training and testing folders named as data. On running it performs necessary data sampling.
2. SOI_HH_Training:
   Here you can see the model description, code to train the model and training logs.
3. SOI_HH_Metric:
   Gives the evaluated accuracy for training and testing samples.
4. SOI_HH_Camera:
   Opens webcam, detects face and shows current emotion.
5. SOI_HH_Submission:
   Used to replicate the submission.csv. example.csv and EMOTOR_TEST folder need to be present inside the repository.