dotnet / machinelearning-modelbuilder

Simple UI tool to build custom machine learning models.
Creative Commons Attribution 4.0 International
265 stars 56 forks source link

Cannot use Object Detection data files created by Azure ML #2460

Open JeremyMahieu opened 1 year ago

JeremyMahieu commented 1 year ago

System Information (please complete the following information):

Describe the bug When selecting an input json file with image labeling, datapreview loads endlessly, can't go to the next step. No relevant info in log files (\AppData\Local\Temp\MLVSTools\logs)

To Reproduce

  1. Add a Machine Learning Model to an empty console project.
  2. Start an "Object detection" scenario
  3. Set up a workspace.
  4. Go to https://ml.azure.com/ and do data labeling on a bunch of images, export the .json
  5. Select the json in the Add data step

Expected behavior Select the .json and be able to go to the next step. Or get an error saying the json is not correct.

Screenshots image image

Example of json file, but really has 40ish images and 100ish annotations.

{
  "images": [
    {
      "id": 1,
      "width": 4032,
      "height": 3024,
      "file_name": "UI/2023-01-28_061527_UTC/asdf/20221213_145249.jpg",
      "coco_url": "AmlDatastore://redacted/UI/2023-01-28_061527_UTC/asdf/20221213_145249.jpg",
      "absolute_url": "https://redacted.blob.core.windows.net/azureml-blobstore-redacted/UI/2023-01-28_061527_UTC/asdf/20221213_145249.jpg",
      "date_captured": "2023-01-28T06:18:25.732655Z"
    }
  ],
  "annotations": [
    {
      "id": 1,
      "category_id": 1,
      "image_id": 1,
      "area": 0.002,
      "bbox": [
        0.27528852681602173,
        0.6812627291242362,
        0.04752206381534285,
        0.033604887983706755
      ]
    }
  ],
  "categories": [
    {
      "id": 1,
      "name": "redacted"
    }
  ]
}
beccamc commented 1 year ago

@JeremyMahieu Sorry you ran into this problem. You used the Azure ML Studio -> Data labelling? Unfortunately we currently only support files generated by Vott.

@luisquintanilla Thoughts on adding this as a data source for OD?

luisquintanilla commented 1 year ago

Makes sense to support the Azure ML format as input. Let's toss it to the backlog for now.

@JakeRadMSFT can correct me here but I believe that's what we do already (though not exposed as an option on input).

JeremyMahieu commented 1 year ago

No I didn't use Vott. Vott has been unmaintained for more than a year, I thought the json files would be compatible. And since the training is on azure etc. At the very least there should be some error saying there's a problem, not just infinite loading. Thanks for the work.

Bjornej commented 1 year ago

Are there any plans to support any formats other than VOTT? The tool has been discontinued and support for other formats could simplify the training ....