将LabelBox JSON输出导入TensorFlow

Question

我正在使用TensorFlow和Keras软件包开发一个对象检测模型。我的第一次迭代没有提供很好的结果，所以我现在使用LabelBox在我的图像周围绘制边界框。LabelBox以如下格式输出一个JSON文件，其中包含所有图像和标签。我正在尝试导入带标签的数据，但找不到一个好的解决方案。我研究过可能需要将JSON转换为TFRecord，但还没有找到一个干净的脚本。我对python和机器学习比较陌生，因此非常感谢您的帮助

LabelBox OutputJson：

{

  "ID":"ck6wf4t9ocmuu0948rgi87ubd",
  "DataRow ID":"ck6wd5n327ea50bofdgbe9tz0",
  "Labeled Data":"https://storage.labelbox.com/ck5x5ecl07i3f0932i797y27k%2Fa9bb10c6-14dd-cc02-1cf6-ccec2b4905e7-Apple_DVI_Adapter_TRAIN_Table_19.jpg?Expires=1585607947466&KeyName=labelbox-assets-key-1&Signature=6Nf7YIr3gjsIkOEZtH2rMy5PJPg",
  "Label":{
     "objects":[
        {
           "featureId":"ck6wf4s5s1ndg0z79nv3x96pf",
           "schemaId":"ck6wdqfe9xqpy0c16ysemtuda",
           "title":"DVI",
           "value":"dvi",
           "color":"#FF0000",
           "bbox":{
              "top":142,
              "left":575,
              "height":694,
              "width":234
           },       "instanceURI":"https://api.labelbox.com/masks/feature/ck6wf4s5s1ndg0z79nv3x96pf?token=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJ1c2VySWQiOiJjazV4NWVjbG9mcXd5MDc3NnEwem9leGpyIiwib3JnYW5pemF0aW9uSWQiOiJjazV4NWVjbDA3aTNmMDkzMmk3OTd5MjdrIiwiaWF0IjoxNTg0Mzk4MzQ3LCJleHAiOjE1ODY5OTAzNDd9.hvNXnIiVNKWfHc_CUONYyXMQWKeY7IqSpqE3z3qNZJQ"
        }
     ],
     "classifications":[

     ]
  },
  "Created By":"",
  "Project Name":"",
  "Created At":"",
  "Updated At":"",
  "Seconds to Label":12.423,
  "External ID":"",
  "Agreement":null,
  "Benchmark Agreement":null,
  "Benchmark ID":null,
  "Benchmark Reference ID":null,
  "Dataset Name":"",
  "Reviews":[

  ],
  "View Label":"https://editor.labelbox.com?project=ck68mszgktd0f0700beuachz5&label=ck6wf4t9ocmuu0948rgi87ubd"

}

Answer 1

我在寻找类似的解决方案。对我来说最有效的是利用Roboflow。我创建了一个免费帐户，能够将LabelBox JSON输出转换为TFRecords，用于模型培训https://roboflow.com/convert/labelbox-json-to-tensorflow-tfrecord