如何准备VOTT JSON数据集以重新训练COCO SSD Tensorflow API？

Question

嗨，我注释了多对象数据集并在VoTT中导出为张量流格式。但是我不知道如何与tensorflow api一起使用它。

Vott为每个带注释的图像生成json文件，如下所示：

"asset": {
    "format": "jpg",
    "id": "0b1e1aac9a6f2cc4e51d95ef368dbfe7",
    "name": "lemon160.jpg",
    "path": "file:/Volumes/Solix/Datasets/limedata/imgs_raw/lemon160.jpg",
    "size": {
        "width": 1280,
        "height": 720
    },
    "state": 2,
    "type": 1
},
"regions": [
    {
        "id": "HWaHAokRV",
        "type": "RECTANGLE",
        "tags": [
            "Expired Lime"
        ],
        "boundingBox": {
            "height": 362.8564453125,
            "width": 510.81555834378923,
            "left": 625.7465495608532,
            "top": 355.5866350446429
        },
        "points": [
            {
                "x": 625.7465495608532,
                "y": 355.5866350446429
            },
            {
                "x": 1136.5621079046425,
                "y": 355.5866350446429
            },
            {
                "x": 1136.5621079046425,
                "y": 718.4430803571429
            },
            {
                "x": 625.7465495608532,
                "y": 718.4430803571429
            }
        ]
    }

}

有人可以建议我将这些文件转换为tfrecords吗？