Overview

Classification and tagging tasks become complex when you deal with large items catalogs (hundreds or thousands of classes). This app transforms labeled project to JSON file with reference items: user has to define reference tag name (objects with this tag will be considered as reference) and key tag name (value of this tag on object is used to group reference items).

For example, in retail labeling there may be thousands of unique UPC codes (Universal Product Code). Labelers can assign tag ref to objects that have to be references. Multiple reference items for the same key (UPC code) are allowed. These reference objects will be grouped by the value of key tag name (e.g. UPC code).

If you already have items catagol we recommend you to convert it into the format described here: it will allow you to use other tagging/classification apps from Ecosystem.

ℹ️ Use the Import Metadata app to add metadata to your images in Supervisely project from JSON file.

How To Use

Step 1: Add app to your team from Ecosystem if it is not there.

Step 2: Run app from the context menu of project

Step 3: Fill in the fields in modal window and press Run button

Step 4: Wait until the task is finished, JSON will be created and saved to team file, find link to it in task output

You can go to (1) Team Files -> (2) /reference_items directory. All JSONs are saved there (3).

Step 5: All warnings and errors can be found in task log

JSON Format

Here is the example:

{
  "project_id": 1219,
  "project_name": "grocery_products",
  "project_url": "http://supervise.ly/projects/1219/datasets",
  "reference_tag_name": "ref",
  "key_tag_name": "UPC",
  "all_keys": ["123", "777"],
  "references": {
    "123": [
      {
        "image_id": 368147,
        "image_name": "IMG_1836.jpeg",
        "dataset_name": "ds1",
        "image_preview_url": "http://supervise.ly/app/images/65/99/1219/1476#image-368147",
        "image_url": "http://supervise.ly/abcd.jpg",
        "UPC": "123",
        "bbox": [122, 569, 273, 706]
      },
      {
        "image_id": 368151,
        "image_name": "IMG_4451.jpeg",
        "dataset_name": "ds1",
        "image_preview_url": "http://supervise.ly/app/images/65/99/1219/1476#image-368151",
        "image_url": "http://supervise.ly/abcd.jpg",
        "UPC": "123",
        "bbox": [100, 249, 286, 421]
      }
    ],
    "777": [
      {
        "image_id": 368150,
        "image_name": "IMG_0748.jpeg",
        "dataset_name": "ds1",
        "image_preview_url": "http://supervise.ly/app/images/65/99/1219/1476#image-368150",
        "image_url": "http://supervise.ly/abcd.jpg",
        "UPC": "777",
        "bbox": [120, 531, 380, 811],
        "geometry": {
          "points": {
            "exterior": [
              [151, 211],
              [296, 390]
            ],
            "interior": []
          },
          "labelerLogin": "max",
          "updatedAt": "2020-11-24T15:16:26.618Z",
          "createdAt": "2020-11-24T15:15:31.892Z",
          "id": 6119695,
          "classId": 18318
        }
      }
    ]
  }
}

Optional top level fields, they are used for information puposes only (just in case):

• project_id (optional) - id of the original project
• project_name (optional) - name of the original project
• project_url (optional) - url to the original project
• reference_tag_name (optional) - this tag is used to find reference objects (in our example we used tag ref)

Mandatory fields:

• key_tag_name - tag that is on every reference object and is used to group them (in our example we used tag UPC)
• all_keys - list of all possible keys (names of the groups) - this array contains all used values of tag defined in key_tag_name field. In our example tag UPC has only two values: 123 and 777
• references - object (in python world it is also named as dictionary), object fields are from all_keys list, and value is the list of reference items

Reference item object (many optional fields are used only for information purposes):

• image_id (optional) - id of the image in Supervisely platform,
• image_name (optional) - name of the image in Supervisely platform,
• dataset_name (optional) - name of the dataset where image is located
• image_preview_url (optional) URL to labeling the image in labeling interface
• image_url - direct url to the image (this url will be used to show and download image in other tagging/classification apps)
• UPC (on your custom case it may differ) - it is a value that is defined in field key_tag_name and the value of this fields is a name of group of reference objects
• bbox - object bounding box (can be an empty list or null) - [top, left, bottom, right] coordinates of the bounding box of the object. if the field is empty - then entire image is used as reference object
• geometry - object in Supervisely JSON format