Emotion Detection Using Yolo-V5 and RepVGG

This repository uses Yolo-V5 and RepVGG to detect facial expressions and classify emotions (see the architecture for more info on how it works). To see how to use the code, check out the usage section for more information.

Example

This is an example of emotion classification: This is a picture of me pulling all 8 of the facial expressions that the model classifies:

Setup

pip

pip install -r requirements.txt

conda

conda env create -f env.yaml

Emotions

This model detects 8 basic facial expressions:

• anger
• contempt
• disgust
• fear
• happy
• neutral
• sad
• surprise
  and then attempts to assign them appropriate colours. It classifies every face, even if it is not that confident about the result!

Usage

usage: main.py [-h] [--source SOURCE] [--img-size IMG_SIZE] [--conf-thres CONF_THRES] [--iou-thres IOU_THRES]
               [--device DEVICE] [--hide-img] [--output-path OUTPUT_PATH | --no-save] [--agnostic-nms] [--augment]
               [--line-thickness LINE_THICKNESS] [--hide-conf] [--show-fps]

optional arguments:
  -h, --help            show this help message and exit
  --source SOURCE       source
  --img-size IMG_SIZE   inference size (pixels)
  --conf-thres CONF_THRES
                        face confidence threshold
  --iou-thres IOU_THRES
                        IOU threshold for NMS
  --device DEVICE       cuda device, i.e. 0 or 0,1,2,3 or cpu
  --hide-img            hide results
  --output-path OUTPUT_PATH
                        save location
  --no-save             do not save images/videos
  --agnostic-nms        class-agnostic NMS
  --augment             augmented inference
  --line-thickness LINE_THICKNESS
                        bounding box thickness (pixels)
  --hide-conf           hide confidences
  --show-fps            print fps to console

Architecture

There are two parts to this code: facial detection and emotion classification.

Face Detection

This repository is a fork of ultralytics/Yolo-V5 because this is the code for classifying faces. Read here for more information on Yolo-V5. To detect faces, the model was trained on the WIDER FACE dataset which has 393,703 faces. For more information, check out the paper here.

Facial Expression Classification

This repository uses code directly from the DingXiaoH/RepVGG repository. You can read the RepVGG paper here to find out more. Even though this is the main model, it made more sense to fork the Yolo-V5 repository because it was more complicated. The model was trained on the AffectNet dataset, which has 420,299 facial expressions. For more information, you can read the paper here.