Detection Offset - Use bbox for percent padding

[sberan]

Description

Padding bounding boxes by percent rather than pixels is a new feature (implemented in https://github.com/roboflow/inference/pull/956/files) and I don't think was implemented correctly. When talking with potential users, they want padding based on a percentage of the size of the bounding box, not the size of the image. This is done because larger detections need more padding because they are closer to the camera. I don't think padding based on image size is very useful. I'm hoping since this functionality is new enough and was implemented for a specific use case that we can modify it.

Type of change

• Bug fix (non-breaking change which fixes an issue)

How has this change been tested, please provide a testcase or example of how you tested the change?

Unit tests cover this change fully.

Any specific deployment considerations

n/a - standard deploy

Docs

• Updated description of input to reflect the use of bounding box percent.