Skip to content

Latest commit

 

History

History

1_Preparation

README.MD

Step 1: Data normalization

This folder contains the source code for image preparation process.

  • Requires Python version 3.7.9
  • OpenCV2 4.7 or newer (pip install opencv-python)
  • Argparse library (pip install argparse)

You can also use the requirements.txt file to install these libraries!

Files

Bounding box management

  • bounding_box_creator.py - Loads files in a folder and allows the user to draw bounding boxes around objects. The output is a json file containing the coordinates of the bounding boxes. Accepts a file with a list of classes for marking! (json file - {"names": [ "class1, "class2" ]}). This is pretty much required if you want to create a dataset for a YOLOv6 or YOLOv8 model.
  • bounding_box_draw.py - A tool to visualize bounding boxes
  • bounding_box_resize.py - A tool to resize images along with their bounding boxes

Dataset preparation

  • prepare_dataset.py - Prepares dataset for SSD and YOLO networks. Please note that SSD preparation only accepts single-class images. For YOLO networks it will create a valid dataset with all of the files marked for training. To further split it, use yolo_data_splitter.py

Image preprocessing

  • process_circledetect.py - Runs circle detection algorithm to cut the plate out of the image. Requires a lot of manual fine-tuning. Does not work well automatically.
  • process_edgedetect.py - Runs edge detection algorithm on the image
  • process_resize.py - Simple image resizer
  • process_subtract.py - Simple image subtractor
  • process_threshold.py - Runs thresholding algorithms on the image

Networks

VGG

  • vgg_file_rename.py - Renames data for the VGG network

YOLO

  • yolo_box_draw.py - Draws an image with boxes on it using YOLO-compatible coordinates
  • yolo_data_splitter.py - Splits data from the training fodler into training, validation and testing datasets. Use this after prepare_dataset.py.
  • yolo_dataset_splitter.py - Splits YOLO dataset into two, in case you need to make your dataset smaller. Will regenerate the data yaml file for you.