Data Preprocessing for Multilabel Classification

Part 1 (2019)
1

Hello everybody,

I am working on an object recognition task with a multi-label classification.
I have picked out about 450 images of construction sites and am now in the process of labeling the images.

Thereby the components are to be classified into one of 3 main classes and into 3 secondary classes. If I label the images with supervis.ly label I get the json file with the following structure.

Blockquote{“description”: “”, “tags”: [], “size”: {“height”: 1836, “width”: 3264},
“objects”: [{“description”: “”, “tags”: [{“name”: “Bauteilzustand”, “value”: “Betoniert”}],
“classTitle”: “Stuetze”, “points”: {“exterior”: [[3264, 0], [3124, 195]], “interior”: []}}]}

Question:

  1. “exterior”: [[3264, 0], [3124, 195]] are the coordinates of the vertices of the rectangle. Do I have to convert the coordinates using Python code or is there another way ?

  2. As mentioned I want to divide the coordinates into exactly 2 classes.
    1 class from the category: wall, column, ceiling
    1 class from the category: concreted, reinforced, shuttering

So the classification should look like this:
Component: Wall
Structural condition: Concreted

Blockquote {“description”: “”, “tags”: [], “size”: {“height”: 1836, “width”: 3264},
“objects”: [{“description”: “”, “tags”: ""
“classTitle”: "Stuetze, Betoniert", “points”: {“exterior”: [[3264, 0], [3124, 195]], “interior”: []}}]}

Does the correct form of my data look like this? And can an object recognition task with multilabel classification be performed with it?

Greetings
Dan