Inconsistent Sharpness Ratings in AWS Rekognition

We are experiencing significant inconsistencies with the sharpness ratings provided by AWS Rekognition. Our workflow involves photographing school events, and we consistently upload high-quality images (2400px on the long edge, JPG format) for processing. Despite this, some of our sharpest images receive low sharpness ratings (as low as 0.2-0.3%), while others receive ratings of 0.6-0.7%.

Could someone provide insights into why these inconsistencies might be occurring and how we can achieve more reliable sharpness ratings?

Steps Taken:

1. We ensured that all images are of high resolution and consistent quality.
2. We checked for variations in lighting and other conditions but found no significant differences that would account for the inconsistency.
3. We experimented with different sets of images and observed similar inconsistent ratings.