dipakbagal / cartoon-emotion-recognition

Tom & Jerry cartoon facial emotion recognition

Cartoon-Emotion-Recognition

Prerequisites

This Repository is maintained on Python 3.7 version.

• CPython 3.6.9
• numpy 1.18.2
• pandas 1.0.3
• torch 1.4.0
• torchvision 0.5.0
• detectron2 0.1.1
• tensorflow 1.15.0

Dataset

Dataset consists of a training video and testing video along with csv file which contains the emotions of the cartoon corresponding to each frame (frame rate = 5) of the video.

Preprocessing

Dataset is in the form of video file from which frames are needed to be extracted. Frames from the video can be obtained using get_frames.py.
The faces for face detection are annotated using labelImg. The annotations for each frame can be transferred to a csv file using xml_to_csv.py.

Dataset for emotion recognition can be obtained using get_dataset.py. It takes a csv file containing the coordinates for the faces and its corresponding emotion and create a dataset for the emotion recognition model.
If you want to train the emotion recognition model on a custom dataset then keep the dataset inside training_dataset images of emotion in a saperate folder.

Training

Face Detection model

The training of face detection model is done in Cartoon_Face_Detection.ipynb. The trained weights can be downloaded from here. Function for prediction on a new image is also present in this notebook.

Emotion Recognition Model

Run the train.sh file to start the training.
Prediction on a new image can be done using get_prediction.py

Outputs