Google Vision API DOCUMENT_TEXT_DETECTION suddenly returns blocks without paragraphs

[jeroenKlippa]

Hello,

We have an issue with  Google Vision API BatchAnnotateImages DOCUMENT_TEXT_DETECTION feature, since March 8, around 6:30PM UTC, we started experiencing issues with the Google Vision API, where it started returning blocks without any paragraphs. Here's some example JSON: 

{
  "fullTextAnnotation": {
    "pages": [
      {
        "blocks": [
          // Other blocks that do have paragraphs.
          {
            "blockType": "TEXT",
            "boundingBox": {
              "vertices": [
                {
                  "x": 739,
                  "y": 985
                },
                {
                  "x": 1548,
                  "y": 985
                },
                {
                  "x": 1548,
                  "y": 1011
                },
                {
                  "x": 739,
                  "y": 1011
                }
              ]
            }
          }
          // A bunch of other blocks that also don't have paragraphs.
          // Other blocks that do have paragraphs. 
        ]
      }
    ]
  }
}

If you look at the demo page of Vision API, it does find the block on the correct position, but no text/paragraph is provided in the block:

It also looks like the demo page itself also doesn't expect blocks without any paragraphs:

Anyone that knows what's going on here? The release notes page doesn't mention any recent release.