This video tutorial walks you through the process of converting binary or labeled masks, often associated with scientific images, into coco style json annotations using python code. It also walks you through the process of converting the coco json annotations to YOLO v8 compatible annotations. As a bonus, I share code to display YOLO v8 annotations using python code.
Code from this video is available here: github.com/bnsreenu/python_fo...
Part 1: Convert binary annotations to COCO JSON
For each binary mask, the code extracts contours using OpenCV.
These contours represent the boundaries of objects within the images. This is a key step in converting binary masks to polygon-like annotations.
Then, convert the contours into annotations, including bounding boxes, area, and segmentation information. Each annotation is associated with an image ID, category ID, and other properties required by the COCO format.
The code also creates an images section containing metadata about the images, such as their filenames, widths, and heights. In my example, I have used exactly the same file names for all images and masks so that a given mask can be easily mapped to the image.
All the annotations, images, and categories are assembled into a dictionary that follows the COCO JSON format. This includes sections for "info," "licenses," "images," "categories," and "annotations."
Finally, the assembled COCO JSON data is saved to a file, making it ready to be used with tools and frameworks that support the COCO data format.
Part 2: Converting COCO JSON annotation to YOLO v8
It reads coco style json annotations supplied as a single json file and also
images as input.
Here are the key steps in the code:
1. Convert Images to YOLO Format: The convert_to_yolo function takes paths for input images and annotations (in JSON format), and directories to store the output images and labels. It then performs the following operations:
- Reads the input JSON file containing annotations.
- Copies all PNG images from the input directory to the output directory.
- Normalizes the polygon segmentation data related to each image and writes them to text files, mapping them to the appropriate category (e.g., Alpha, Cells, Mito, Vessels).
- The resulting text files contain information about the object category and the normalized coordinates of the polygons that describe the objects.
2. Create YAML Configuration File: The create_yaml function takes paths to the input JSON file containing categories, training, validation, and optional test paths. It then:
- Extracts the category names and the number of classes.
- Constructs a dictionary containing information about class names, the number of classes, and paths to the training, validation, and test datasets.
- Writes this dictionary to a YAML file, which can be used as a configuration file for training a model (e.g., a YOLO model).
19 сен 2023
