Repo for the SUFST driverless cone detection model

└─ main.py          - Home screen gui, ML model controller, feature definition
└─ train.py         - ML model training script
└─ data.yaml        - Data file containing key ML model parameters
└─ requirements.txt - Python package/library requirements
└─ data
  └─ images
    └─ train        - Folder for training images
    └─ val          - Folder for validation images
  └─ lables
    └─ train        - Folder for training labels
    └─ val          - Folder for validation labels

• Install git
  • https://git-scm.com/download/win
  • Ensure git is added to PATH
• Clone git repo locally:

git clone https://github.com/sufst/DV.ConeDetection.git

• Make sure all dependencies are installed:

pip install -r requirements.txt

• Run main.py to launch the UI

• 'Insert Image'
  • Allows selection of locally stored image into either train or validation datasets
• 'Sample Video'
  • Allows sampling of locally stored videos into either train or validation datasets
• 'Labelled Unlabelled'
  • Opens the annotations tool for all currently unlabelled images
• 'Redraw All'
  • Opens the annotations tool for all currently stored images
  • Overwrites current annotations/labels
• 'Train Model'
  • Opens the model training page, allowing you to start model training with the current database
  • Live view of all model training outputs

• 'Image'
  • Image name, can be used to find image file in the images directory
• 'Set'
  • What data set the image is applied to
  • Either 'train' or 'val' (validation)
• 'Status
  • What is the annotation status of the image
  • Either 'Labelled' or 'Unlabelled'
• 'Redraw'
  • Opens the annotation tool for that image only
  • Overwrites that image's current annotations/labels
• 'Visualise'
  • Displays that images current annotations/labels
• 'Transfer'
  • Transfers the current data set the image is part of
  • Eg. from 'train' to 'val'
• 'Delete'
  • Deletes image from the dataset

The annotator tool is used for labeling training data, classifying defined regoins, which the ML model uses for object detection.

The annotator tool can be opened by:

• Clicking the 'Redraw' button on the home page
• Automatically opens when you sample a video
  • This has the added feature of a slider which allows you to scrub to specific frames

Labels can be drawn by clicking and dragging, this creates a box drawn on the screen.

• 0 - Cone label selection
• 1 - Non-cone label selection
• S - Save current labels/annotations for thta image
• Q - Quit or close the annotation viewer and return to the home page