Data Aggregation

Once the data is generated, the data can be aggregated into specific spec formats that can later be used for tasks such as ingestion to data curation tools or fine-tuning of a pre-trained model. The formats we support are:

• COCO JSON

• JSON

• JSONL

• CSV

Below are the structures of these specs for reference:

COCO JSON

Note - This spec only supports segmentation data (when segmentation data collection is enabled). None of the other specs hold segmentation data

{
    "info": {
        "description": "DFW Takeoff-Landing Data",
        "url": "",
        "version": 1.0,
        "year": 2021,
        "contributor": "Project AirSim"
    },
    "licenses": [
        {
            "id": 0,
            "url": "/",
            "name": "Public Domain"
        }
    ],
    "images": [
        {
            "license": 0,
            "file_name": "images/0.png",
            "coco_url": "{Link to image on AML datastore}", 
            "absolute-url": "{Link to image on blob-container}",
            "height": 225,
            "width": 400,
            "id": "0",
            "date-captured": ""
        }
    ],
    "categories": [
        {
            "supercategory": "none",
            "id": 0,
            "name": "LandingPads"
        }
    ],
    "annotations": [
        {
            "segmentation": [],
            "iscrowd": 0,
            "area": 52200,
            "image_id": "0",
            "bbox": [
                0.21,
                0.0,
                0.58,
                1.0
            ],
            "category_id": 0,
            "id": "0"
        }
    ]
}

JSON

{
    "baseuri": "{blob-container URL}",
    "categories": [
        "BasicLandingPad"
    ],
    "dataset_name": "DatasetTwoTestTwo",
    "images": [
        {
            "image-width": 400,
            "image-height": 225,
            "file": {
                "hash": "",
                "key": "",
                "object-type": "File",
                "storage-type": "azure-blob",
                "uri": "0.png"
            },
            "format": ".png",
            "split": "validate",
            "polygon": [
               {"3d BBOX Data"}
            ],
            "boxes": [
                {
                    "height": 0.1111111111111111,
                    "left": 0.4375,
                    "score": 1.0,
                    "tag": "AirTaxi",
                    "top": 0.2088888888888889,
                    "width": 0.1225
                }
            ],
            "weather": "RAIN",
            "time": "2022-06-20T07:15:00",
            "geo-location": "DFW",
            "lat-lon": "33.032079,-97.284227"
        }
    ]
}

JSONL

{"image_url": "{Link to image}", "image-width": 400, "image-height": 225, "label": {"label": "AirTaxi", "isCrowd": false, "label_confidence": [1.0], "topX": 0.4375, "topY": 0.2088888888888889, "bottomX": 0.56, "bottomY": 0.32, "polygon": [[217.0, 72.0], [218.0, 57.0], [182.0, 72.0], [181.0, 57.0], [223.0, 67.0], [224.0, 47.0], [176.0, 67.0], [175.0, 47.0]]}},

CSV

ImageName	pose_x	pose_y	pose_z	roll	pitch	yaw	category	x0	y0	x1	y1	x2	y2	x3	y3	x4	y4	x5	y5	x6	y6	x7	y7	x_c	y_c	w	h
0.png	-557	5	-154	0	0	3.141592654	AirTaxi	217	72	218	57	182	72	181	57	223	67	224	47	176	67	175	47	199.5	59.5	49	25
1.png	-557.04	5	-154	0	0	3.141592654	AirTaxi	217	72	218	57	182	72	181	57	223	67	224	46	176	67	175	46	199.5	59	49	26