Google Cloud Vision- Method: files.annotate - Response Object Changing - google-cloud-platform

Question

As per the documentation provided in Cloud Vision Docs the BoundingPoly object in the blocks array should have a format like this
{
"vertices": [
{
object (Vertex)
}
],
"normalizedVertices": [
{
object (NormalizedVertex)
}
]
}
But when we tried https://vision.googleapis.com/v1/files:annotate?key=xxxxxx to perform OCR on a PDF file with the request:
{
"requests": [{
"inputConfig": {
"content": "encoded content",
"mimeType": "application/pdf"
},
"features": [{
"type": "DOCUMENT_TEXT_DETECTION",
"maxResults": 50
}]
}]
}
the response from server was
{
"responses": [
{
"responses": [
{
"fullTextAnnotation": {
"pages": [
{
"property": {
"detectedLanguages": [
{
"languageCode": "en",
"confidence": 0.65
},
{
"languageCode": "fil",
"confidence": 0.01
}
]
},
"width": 841,
"height": 595,
"blocks": [
{
"property": {
"detectedLanguages": [
{
"languageCode": "en",
"confidence": 1
}
]
},
"boundingBox": {
"normalizedVertices": [
{
"x": 0.4351962,
"y": 0.057142857
},
{
"x": 0.6052319,
"y": 0.057142857
},
{
"x": 0.6052319,
"y": 0.08571429
},
{
"x": 0.4351962,
"y": 0.08571429
}
]
},
"paragraphs": [
{
"property": {
"detectedLanguages": [
{
"languageCode": "en",
"confidence": 1
}
]
},
"boundingBox": {
"normalizedVertices": [
{
"x": 0.4351962,
"y": 0.057142857
},
{
"x": 0.6052319,
"y": 0.057142857
},
{
"x": 0.6052319,
"y": 0.08571429
},
{
"x": 0.4351962,
"y": 0.08571429
}
]
},
"words": [
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
]
},
"boundingBox": {
"normalizedVertices": [
{
"x": 0.4351962,
"y": 0.057142857
},
{
"x": 0.49346018,
"y": 0.057142857
},
{
"x": 0.49346018,
"y": 0.08571429
},
{
"x": 0.4351962,
"y": 0.08571429
}
]
},
"symbols": [
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
]
},
"text": "F",
"confidence": 0.99
},
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
]
},
"text": "a",
"confidence": 1
},
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
]
},
"text": "c",
"confidence": 0.99
},
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
]
},
"text": "t",
"confidence": 0.99
},
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
]
},
"text": "o",
"confidence": 1
},
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
]
},
"text": "r",
"confidence": 1
},
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
],
"detectedBreak": {
"type": "SPACE"
}
},
"text": "y",
"confidence": 1
}
],
"confidence": 0.99
},
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
]
},
"text": "i",
"confidence": 0.99
},
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
]
},
"text": "n",
"confidence": 1
},
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
],
"detectedBreak": {
"type": "SPACE"
}
},
"text": "g",
"confidence": 1
}
],
"confidence": 0.99
},
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
]
},
"boundingBox": {
"normalizedVertices": [
{
"x": 0.57431626,
"y": 0.057142857
},
{
"x": 0.6052319,
"y": 0.057142857
},
{
"x": 0.6052319,
"y": 0.08571429
},
{
"x": 0.57431626,
"y": 0.08571429
}
]
},
"symbols": [
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
]
},
"text": "L",
"confidence": 0.99
},
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
]
},
"text": "i",
"confidence": 0.99
},
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
]
},
"text": "s",
"confidence": 0.99
},
{
"property": {
"detectedLanguages": [
{
"languageCode": "en"
}
],
"detectedBreak": {
"type": "LINE_BREAK"
}
},
"text": "t",
"confidence": 1
}
],
"confidence": 0.99
}
],
"confidence": 0.99
}
],
"blockType": "TEXT",
"confidence": 0.99
}
Is there anything to be considered if the vertices object is missing in the BoundingPoly object(boundingBox) property in the above json
When tried in the drag & drop demo, the json response for the OCR done on an image was
"fullTextAnnotation": {
"pages": [
{
"blocks": [
{
"blockType": "TEXT",
"boundingBox": {
"vertices": [
{
"x": 31,
"y": 63
},
{
"x": 147,
"y": 63
},
{
"x": 147,
"y": 81
},
{
"x": 31,
"y": 81
}
]
},
"confidence": 0.99,
"paragraphs": [
{
"boundingBox": {
"vertices": [
{
"x": 31,
"y": 63
},
{
"x": 147,
"y": 63
},
{
"x": 147,
"y": 81
},
{
"x": 31,
"y": 81
}
]
},
Is this the intended behavior or any issues ? which field we should follow normalizedVertices or vertices !!

Answer 1

The difference is that, in the request made from the code you're sending a PDF. In the drag & drop demo, you're sending an image (the demo doesn't accept files).
I replicated this and the behavior seems to be constant: PDF files are annotated with NormalizedVertices, whereas images are annotated with Vertices. My guess is that this is the intended behavior to enhance the performance of large PDF files annotation requests (large due to number of pages).
I sent a request to the Google Documentation so they can add this information in their docs.