IssueTracker

Issue summary: - Customer is having multiple issues with OCR (document_text_detection)

Difference in output when using different values for batchSize parameter. Tested with batchSize=1 and the output seems good while using batchSize=8 produced a very small output.

Some letters in the PDF does not match the output (OCR output quality)

Ex. సహకాసోదరునితో should be సహ=కాసోదరునితో

The sequence of words appearing is in incorrect order as compared to the actual PDF (OCR output quality)

Some characters are missing (OCR output quality)

Customer is looking/following this documentation [1]
See attached PDF file (AK-Sarga1-part.pdf) which contains "TELUGU" language.
[1] https://cloud.google.com/vision/docs/reference/rest/v1/files/asyncBatchAnnotate

Reproduction Steps: Used this command:

curl -X POST -H "Authorization: Bearer <AUTH_TOKEN>" -H "Content-Type: application/json; charset=utf-8" https://vision.googleapis.com/v1/files:asyncBatchAnnotate -d @<path to json request files>

With the following config:

{
"requests": [
{
"features": [
{
"type": "DOCUMENT_TEXT_DETECTION"
}
],
"inputConfig": {
"gcsSource": {
"uri": "[BUCKET]/AK-Sarga1-part.pdf"
},
"mimeType": "application/pdf"
},
"imageContext": {
"languageHints": [
"te"
],
"textDetectionParams": {
"enableTextDetectionConfidenceScore": false
}
},
"outputConfig": {
"batchSize": 8,
"gcsDestination": {
"uri": "gs://sriramayanamfiles/pdf-test/"
}
}
}
]
}