How can one convert a customized JSON dataset file into YOLOv8 text file format using Python?

edit

To convert a customized JSON dataset file into YOLOv8 text file format using Python, you can follow the below steps:

1. Load the JSON file and extract the class labels, image dimensions, and object details using the json library.

2. Iterate through each object in the JSON file and extract the object's class, coordinates, height, and width.

3. Map each object's class label to a unique integer index.

4. Normalize the object coordinates and dimensions to values between 0 and 1, using the image dimensions.

5. Write each object's details to a text file in the YOLOv8 format, which consists of one line per object: class index, normalized x-coordinate, normalized y-coordinate, normalized width, and normalized height.

Here is an example code snippet that performs the above conversion:

import json

# Load JSON file
with open('custom_dataset.json') as f:
    json_data = json.load(f)

# Extract class labels and image dimensions
class_labels = json_data['class_labels']
img_height = json_data['img_height']
img_width = json_data['img_width']

# Create class label to index mapping
class_index = {class_labels[i]: i for i in range(len(class_labels))}

# Write object details to YOLOv8 formatted text file
with open('yolo_dataset.txt', 'w') as f:
    for obj in json_data['objects']:
        # Extract object details
        obj_class = obj['class']
        obj_x = obj['x']
        obj_y = obj['y']
        obj_width = obj['width']
        obj_height = obj['height']

        # Normalize coordinates and dimensions
        normalized_x = obj_x / img_width
        normalized_y = obj_y / img_height
        normalized_width = obj_width / img_width
        normalized_height = obj_height / img_height

        # Write to text file in YOLOv8 format
        f.write(f"{class_index[obj_class]} {normalized_x} {normalized_y} {normalized_width} {normalized_height}\n")

Note: The above code assumes that the JSON file contains an array of object details, where each object contains the class label, coordinates, height, and width. The code may need to be modified depending on the JSON structure in your dataset file.