A.Eye

한국 데이터 산업진흥원에서 주최한 데이터 청년 캠퍼스

고려대학교 소속 13조입니다.

This is Korea University Team 13 of Data Campus hosted by Korea Data Agency.

Directory Structure

.
├── cls_datasets        # image data for train and validation
│   ├── airpods         # airpods data
│   ├── book            # book data
│   ├── cellphone       # cell phone data
│   ├── desklamp        # desk lamp data
│   ├── keyboard        # keyboard data
│   ├── laptop          # laptop data
│   ├── mouse           # mouse data
│   ├── pen             # pen data
│   ├── pencilcase      # pencilcase data
│   ├── stopwatch       # stopwatch data
│   ├── tablet computer # tablet computer data
│   └── wallet          # wallet data
├── data_preprocessing  # codes for data preprocessing
├── project             # our main project
│   ├── data            # .ymal data
│   ├── models          # yolo model
│   └──  utils          # other codes
└──  youtube_dataset    # image data from youtube (vd1~vd10)
    ├── vd6             # images and annotations
    ├── vd7             # images and annotations
    └── vd9             # images and annotations

Demo in colab

# You can clone github
!git clone https://github.com/carrytheeast/Team13

# Go to code folder
%cd Team13/project

# Download weights. If not working, You can try click to download release on this web. 
!wget https://github.com/carrytheeast/Team13/releases/download/v0.1/best.pt

# pip install required packages
!pip install -r requirements.txt

Inference

On video:

# detect.py
# --mode 0 for user (default)
# --mode 1 for developer
!python detect.py --weights best.pt --source [your_video.mp4] --save-path [your/path/name.mp4] --mode 0

Use Viewer in Local(Quickly Visualization)

##### Do not working in Colab #####
# viewer.py
!python viewer.py --weight best.pt --source [yourvideo.mp4] --save-path [your/path/name.mp4] --mode 0 

Woking

No Detect Face and Eyes:

Draw guide lines.

User Mode

On Working:

User can check the belows.

• pure study time

• total time

When EAR is CLOSE, We show the sentence at middle. (1)

Developer Mode

On Working:

Developer can check the belows.

• Gaze Directions
• Gaze Lines
• Selected Area
• Objects Labels
• Object Bboxes
• Time

On Working:

When EAR is CLOSE, We show the sentence at middle. (2)

Result Visualtization

After detect, Save run.mp4 and fig1.png

detec.py options

detec.py supports the following options:

usage: detect.py  [-h] [--weights WEIGHTS [WEIGHTS ...]] [--source SOURCE]
                  [--img-size IMG_SIZE] [--conf-thres CONF_THRES]
                  [--iou-thres IOU_THRES] [--device DEVICE] [--view-img]
                  [--classes CLASSES [CLASSES ...]] [--agnostic-nms]
                  [--augment] [--update] [--no-trace] [--save-path SAVE_PATH]
                  [--mode MODE]

References

• https://github.com/WongKinYiu/yolov7
• https://github.com/google/mediapipe
• https://github.com/albumentations-team/albumentations