Coco Json file to CSV format (path/to/image.jpg,x1,y1,x2,y2,class_name)

Question

I would like to convert my coco JSON file as follows:

The CSV file with annotations should contain one annotation per line. Images with multiple bounding boxes should use one row per bounding box. Note that indexing for pixel values starts at 0. The expected format of each line is:

path/to/image.jpg,x1,y1,x2,y2,class_name

A full example:

*/data/imgs/img_001.jpg,837,346,981,456,cow 
/data/imgs/img_002.jpg,215,312,279,391,cat
/data/imgs/img_002.jpg,22,5,89,84,bird

This defines a dataset with 3 images: img_001.jpg contains a cow, img_002.jpg contains a cat and a bird, and img_003.jpg contains no interesting objects/animals.

How could I do that?

Answer 1

I have such function.

def convert_coco_json_to_csv(filename):
    import pandas as pd
    import json
    
    # COCO2017/annotations/instances_val2017.json
    s = json.load(open(filename, 'r'))
    out_file = filename[:-5] + '.csv'
    out = open(out_file, 'w')
    out.write('id,x1,y1,x2,y2,label\n')

    all_ids = []
    for im in s['images']:
        all_ids.append(im['id'])

    all_ids_ann = []
    for ann in s['annotations']:
        image_id = ann['image_id']
        all_ids_ann.append(image_id)
        x1 = ann['bbox'][0]
        x2 = ann['bbox'][0] + ann['bbox'][2]
        y1 = ann['bbox'][1]
        y2 = ann['bbox'][1] + ann['bbox'][3]
        label = ann['category_id']
        out.write('{},{},{},{},{},{}\n'.format(image_id, x1, y1, x2, y2, label))

    all_ids = set(all_ids)
    all_ids_ann = set(all_ids_ann)
    no_annotations = list(all_ids - all_ids_ann)
    # Output images without any annotations
    for image_id in no_annotations:
        out.write('{},{},{},{},{},{}\n'.format(image_id, -1, -1, -1, -1, -1))
    out.close()

    # Sort file by image id
    s1 = pd.read_csv(out_file)
    s1.sort_values('id', inplace=True)
    s1.to_csv(out_file, index=False)

Answer 2

Here is a function I use to convert Coco format to AutoML CSV format for image object detection annotated data:

def convert_coco_json_to_csv(filename,bucket):
    import pandas as pd
    import json
    
    s = json.load(open(filename, 'r'))
    out_file = filename[:-5] + '.csv'

    with open(out_file, 'w') as out:
      out.write('GCS_FILE_PATH,label,X_MIN,Y_MIN,,,X_MAX,Y_MAX,,\n')
      file_names = [f"{bucket}/{image['file_name']}" for image in s['images']]
      categories = [cat['name'] for cat in s['categories']]
      for label in s['annotations']:
        #The COCO bounding box format is [top left x position, top left y position, width, height]. 
        # for AutoML: For example, a bounding box for the entire image is expressed as (0.0,0.0,,,1.0,1.0,,), or (0.0,0.0,1.0,0.0,1.0,1.0,0.0,1.0).
        HEIGHT = s['images'][label['image_id']]['height']
        WIDTH = s['images'][label['image_id']]['width']
        X_MIN = label['bbox'][0]/WIDTH
        X_MAX = (label['bbox'][0] + label['bbox'][2]) / WIDTH
        Y_MIN = label['bbox'][1] / HEIGHT
        Y_MAX = (label['bbox'][1] + label['bbox'][3]) / HEIGHT
        out.write(f"{file_names[label['image_id']]},{categories[label['category_id']]},{X_MIN},{Y_MIN},,,{X_MAX},{Y_MAX},,\n")

And simply you can use it by calling the function with the file name and the gs storage where images were uploaded:

convert_coco_json_to_csv("/content/train_annotations.coco.json", "gs://[bucket name]")

Answer 3

Posting my function in case someone finds it useful after exporting from Label Studio (AutoML export format is currently not supported) and wants to use it with TFlite Model Maker.

def convert_coco_json_to_csv(filename, labels):
    import json, random
    
    s = json.load(open(filename, 'r'))
    
    # Remember image paths by id
    images = {}
    for im in s['images']:
        images[im['id']] = {
            'path': im['file_name'].split('/')[-1], # Split likely not required in most cases
            'width': im['width'],
            'height': im['height']
        }

    images = list(images.items()) # Cannot shuffle a dictionary
    random.shuffle(images)
    images = dict(images)
    nr_of_annotations = len(s['annotations']) - 1

    # Write to Google Cloud AutoML format .csv
    out_file = filename[:-5] + '.csv'
    out = open(out_file, 'w')

    # set,path,label,x_min,y_min,,,x_max,y_max,,
    out.write('set,path,label,x_min,y_min,,,x_max,y_max,,\n')

    for i, ann in enumerate(s['annotations']):
        x_min = ann['bbox'][0] / images[ann['image_id']]['width']
        x_max = (ann['bbox'][0] + ann['bbox'][2]) / images[ann['image_id']]['width']
        y_min = ann['bbox'][1] / images[ann['image_id']]['height']
        y_max = (ann['bbox'][1] + ann['bbox'][3]) / images[ann['image_id']]['height']

        # Split images into train, validation and test sets by 75%, 20% and 5% respectively
        percentage = i / nr_of_annotations * 100
        if percentage < 75:
            img_set = 'TRAIN'
        elif percentage < 95:
            img_set = 'VALIDATION'
        else:
            img_set = 'TEST'

        path = images[ann['image_id']]['path']
        label = labels[int(ann['category_id'])]
        out.write('{},{},{},{},{},{},{},{},{},{},{}\n'.format(img_set, path, label, x_min, y_min, '', '', x_max, y_max, '', ''))
    out.close()