AWS Rekognition - creating a dataset with a manifest file is always throwing an error

Question

I am creating a Dataset for a project in AWS rekognition by using a SageMaker formatted manifest file as below :

{
  "source-ref": "s3://custom-labels-console-ap-south-1-ddd62b2581/assets/pwst-new/1706007842/dot.jpg",
  "pwst-new-train_BB": {
    "annotations": [
      {
        "left": 99,
        "top": 66,
        "width": 202,
        "height": 335,
        "class_id": 0
      }
    ],
    "image_size": [
      {
        "width": 640,
        "height": 480,
        "depth": 3
      }
    ]
  },
  "pwst-new-train_BB-metadata": {
    "job-name": "labeling-job/pwst-new-train_BB",
    "class-map": {
      "0": "test"
    },
    "human-annotated": "yes",
    "objects": [
      {
        "confidence": 1
      }
    ],
    "creation-date": "2024-01-23T11:04:35.981Z",
    "type": "groundtruth/object-detection"
  },
  "cl-metadata": {
    "is_labeled": true
  }
}

I have added this file to an S3 bucket with correct Policy and still when I am trying to create a dataset, it's always throwing The manifest file contains too many invalid rows.

I tried adding the manifest file to different folders and also in the bucket that was created by default when launching Rekognition for the first time.

I also tried to just copy paste the contents of

aws rekognition list-dataset-entries

after creating a dataset by manually drawing blocks around directly uploaded images