How to Convert COCO Json to YOLOv8 segmentation format

def convert_coco_to_yolov8(coco_file):

with open(coco_file) as f:
    coco = json.load(f)


images = coco['images']
annotations = coco['annotations'] 
categories = {cat['id']: cat['name'] for cat in coco['categories']}


os.makedirs('labels', exist_ok=True)


for image in tqdm(images, desc='Converting images'):
    image_id = image['id']
    filename = image['file_name']



    label_filename = filename.split('.png')[0]
    label_path = os.path.join('labels', f'{label_filename}.txt')
    with open(label_path, 'w') as f:

        for ann in annotations:
            if ann['image_id'] != image_id:
                continue

            img_width = image['width']
            img_height = image['height']

            xmin, ymin, width, height = ann['bbox']
          
            xmax = xmin + width
            ymax = ymin + height
            xcen = (xmin + xmax) / 2
            ycen = (ymin + ymax) / 2
            # xcen = (xmin + xmax) / 2 / img_width
            # ycen = (ymin + ymax) / 2 / img_height
            w = xmax - xmin
            h = ymax - ymin
            label = categories[ann['category_id']]
            label_id = ann['category_id']
         
        

            segmentation_points_list = []
            for segmentation in ann['segmentation']:
                segmentation_points = [str(point / img_width) for point in segmentation]
                segmentation_points_list.append(' '.join(segmentation_points))
            segmentation_points_string = ' '.join(segmentation_points_list)


                            

         
            line = '{} {} {} {} {} {}\n'.format(label_id, xcen / img_width, ycen / img_height, w / img_width, h / img_height, segmentation_points_string )
            f.write(line)

the script is getting the labels but when i train for YOLOv8 the labels are seems wrong ,I need to convert a coco json to YOLOV8 txt file . label should contain segmentation also. Note my JSON file have different image size for all images

This works for me

I have loaded all the images in images/all_images directory. I used coco .json annotations differently for train/test/val. You can change accordingly. Make sure to create directory yourself.

<!-- begin snippet: js hide: false console: true babel: false -->

<!-- begin snippet: js hide: false console: true babel: false -->

<!-- language: lang-html -->

import os
import json
import shutil
# load json and save directory for labels train/val/test
coco_file = 'labels/val.json'
save_folder = 'labels/val'
# 0 for truck, 1 for car
category = 1
#source of all the images and destination folder for train/test/val
source_path = "images/all_images"
destination_path = "images/val"
# Use os.listdir() to get a list of filenames in the folder
file_names = os.listdir(source_path)
with open(coco_file) as f:
coco = json.load(f)
images = coco['images']
annotations = coco['annotations'] 
categories = {cat['id']: cat['name'] for cat in coco['categories']}
#os.makedirs(save_folder, exist_ok=True)
for ann in annotations:
image = next(img for img in images if (img['id'] == ann['image_id']))
if (image["file_name"] not in file_names):
continue
width, height = image['width'], image['height']
x_coco, y_coco, width_coco, height_coco = ann['bbox']
category_id = ann['category_id']
image_id = ann['image_id']
filename = image['file_name']
label_filename = filename.split('.jpg')[0]
label_path = os.path.join(save_folder, f'{label_filename}.txt')
with open(label_path, 'a') as f:
segmentation_points_list = []
for segmentation in ann['segmentation']:
# Check if any element in segmentation is a string
if any(isinstance(point, str) for point in segmentation):
continue  # Skip this segmentation if it contains strings
segmentation_points = [str(float(point) / (width-1) if i % 2 == 0 else float(point) / (height-1)) for i, point in enumerate(segmentation)]
segmentation_points_list.append(' '.join(segmentation_points))
segmentation_points_string = ' '.join(segmentation_points_list)
line = '{} {}\n'.format(category, segmentation_points_string)
f.write(line)
segmentation_points_list.clear()
image_source = source_path + f'/{image["file_name"]}'
shutil.copy(image_source, destination_path)

<!-- end snippet -->

<!-- language: lang-html -->

import os
import json
import shutil
# load json and save directory for labels train/val/test
coco_file = 'labels/val.json'
save_folder = 'labels/val'
#source of all the images and destination folder for train/test/val
source_path = "images/all_images"
destination_path = "images/val"
# Use os.listdir() to get a list of filenames in the folder
file_names = os.listdir(source_path)
with open(coco_file) as f:
coco = json.load(f)
images = coco['images']
annotations = coco['annotations'] 
categories = {cat['id']: cat['name'] for cat in coco['categories']}
#os.makedirs(save_folder, exist_ok=True)
for ann in annotations:
image = next(img for img in images if (img['id'] == ann['image_id']))
if (image["file_name"] not in file_names):
continue
#print(f"image in annotations =   {type(image['id'])}")
width, height = image['width'], image['height']
x_center = (ann['bbox'][0] + ann['bbox'][2] / 2) / width
y_center = (ann['bbox'][1] + ann['bbox'][3] / 2) / height
bbox_width = ann['bbox'][2] / width
bbox_height = ann['bbox'][3] / height
category_id = ann['category_id']
image_id = ann['image_id']
filename = image['file_name']
label_filename = filename.split('.jpg')[0]
label_path = os.path.join(save_folder, f'{label_filename}.txt')
with open(label_path, 'a') as f:
segmentation_points_list = []
for segmentation in ann['segmentation']:
# Check if any element in segmentation is a string
if any(isinstance(point, str) for point in segmentation):
continue  # Skip this segmentation if it contains strings
segmentation_points = [str(float(point) / width) for point in segmentation]
segmentation_points_list.append(' '.join(segmentation_points))
segmentation_points_string = ' '.join(segmentation_points_list)
line = '{} {}\n'.format(categories, segmentation_points_string)
f.write(line)
segmentation_points_list.clear()
image_source = source_path + f'/{image["file_name"]}'
shutil.copy(image_source, destination_path)

<!-- end snippet -->

Look at the label format used in this script:

line = '{} {} {} {} {} {}'.format(label_id, x_center, y_center, w_box, h_box, segmentation_points_str)

For instance segmentation format use this: class x1 y1 x2 y2 ... xn yn. So I recommend changing this line to:

line = '{} {}'.format(label_id, segmentation_points_str)

Instance segmentation label format: https://docs.ultralytics.com/datasets/segment/