Custom Vision Service

  • Out of the box APIs:
  • Custom Vision
    • There are 2 types of projects you can create:
      • Object Recognition
        • Here, you can train your Vision AI to find checkboxes in a doc:
          • Don’t use the following training data pattern:
            • trainCB.PNG
          • Instead, the training data should be real docs:
            • trainR1.PNG
          • Then manually mark checkboxes as regions:
            • trainR2.PNG
        • Finally test the Model using the real Image like so:
          • doc.PNG
        • Here’s the accuracy I see with only 15 images in training set:
          • 1 region matched
      • Classification
    • Export the project’s model:

Hyperledger Explorer