COCO JSON

COCO is a common JSON format used for machine learning because the dataset it was introduced with has become a common benchmark.

Overview

Microsoft released the MS COCO dataset in 2015. It has become a common benchmark dataset for object detection models since then which has popularized the use of its JSON annotation format. You can learn how to create COCO JSON from scratch in our CVAT tutorial.

Unfortunately, COCO format is not anywhere near universal and so you may find yourself needing to convert it to another format for a model (or export to COCO JSON from another format if you happen to be using a model that supports it).

Not to worry, Roboflow is the universal tool for computer vision format conversion and can seamlessly input and output files in COCO JSON format.

Format Description

Below, learn the structure of COCO JSON.

coco.jsonJSON

{
    "info": {
        "year": "2020",
        "version": "1",
        "description": "Exported from roboflow.ai",
        "contributor": "Roboflow",
        "url": "https://app.roboflow.ai/datasets/hard-hat-sample/1",
        "date_created": "2000-01-01T00:00:00+00:00"
    },
    "licenses": [
        {
            "id": 1,
            "url": "https://creativecommons.org/publicdomain/zero/1.0/",
            "name": "Public Domain"
        }
    ],
    "categories": [
        {
            "id": 0,
            "name": "Workers",
            "supercategory": "none"
        },
        {
            "id": 1,
            "name": "head",
            "supercategory": "Workers"
        },
        {
            "id": 2,
            "name": "helmet",
            "supercategory": "Workers"
        },
        {
            "id": 3,
            "name": "person",
            "supercategory": "Workers"
        }
    ],
    "images": [
        {
            "id": 0,
            "license": 1,
            "file_name": "0001.jpg",
            "height": 275,
            "width": 490,
            "date_captured": "2020-07-20T19:39:26+00:00"
        }
    ],
    "annotations": [
        {
            "id": 0,
            "image_id": 0,
            "category_id": 2,
            "bbox": [
                45,
                2,
                85,
                85
            ],
            "area": 7225,
            "segmentation": [],
            "iscrowd": 0
        },
        {
            "id": 1,
            "image_id": 0,
            "category_id": 2,
            "bbox": [
                324,
                29,
                72,
                81
            ],
            "area": 5832,
            "segmentation": [],
            "iscrowd": 0
        }
    ]
}

Split and Merge Datasets

With Roboflow supervision, an open source Python package with utilities for completing computer vision tasks, you can merge and split detections in COCO JSON. Read our dedicated guides to learn how to merge and split COCO JSON detections.

Split detections in COCO JSON

Python

import supervision as sv

ds = sv.DetectionDataset.from_coco(
    images_directory_path=f"dataset1/train",
    annotations_path=f"dataset1/train/_annotations.coco.json",
)

train_ds, test_ds = ds.split(split_ratio=0.7,
                             random_state=42, shuffle=True)
len(train_ds), len(test_ds)
# (700, 300)

Supported Models

Below, see model architectures that require data in the COCO JSON format when training a new model.

YOLOv8 Pose EstimationKeypoint Detection
YOLOv8 Pose Estimation predicts 17-keypoint human skeletons in real time using the YOLOv8 architecture, deployable via Roboflow Inference on custom datasets.
DETRObject Detection
DETR is Meta's end-to-end transformer-based object detector that eliminates anchors and NMS, achieving 42.0% AP on COCO using set prediction with attention.
YOLOSObject Detection
YOLOS applies a Vision Transformer to object detection as a flat sequence without convolutions or anchors, demonstrating ViT transferability to detection tasks.
Mask R-CNNInstance Segmentation
Mask R-CNN extends Faster R-CNN with a mask prediction branch and RoIAlign, enabling precise instance segmentation alongside object detection in a single model.
EfficientDetObject Detection
EfficientDet is Google's scalable object detection model using compound scaling and BiFPN, achieving 55.1% AP on COCO across a family of model sizes.
Detectron2Object Detection
Detectron2 is Meta's PyTorch-based computer vision framework supporting object detection, instance segmentation, keypoint detection, and panoptic segmentation.

COCO JSON

Overview

Format Description

Split and Merge Datasets

Split detections in COCO JSON

Supported Models

Convert Data from COCO JSON

Convert Data to COCO JSON

Speak with an AI expert