Feat: allow bounding box transform functions to take in multi-dimensional inputs

aisingapore / PeekingDuck

A modular framework built to simplify Computer Vision inference workloads.

Apache License 2.0

163 stars 39 forks source link

Feat: allow bounding box transform functions to take in multi-dimensional inputs #711

Closed gao-hongnan closed 2 years ago

gao-hongnan commented 2 years ago

This pull request addresses the issue at https://github.com/aisingapore/PeekingDuck-Private/issues/97.

Changed all functions in utils.bbox.transforms to allow multi-dimensional input by using the ellipsis operator [..., ].
Added helper functions to clone all inputs and cast them as float. The former will ensure no mutation to original inputs whereas the latter ensures the outputs will be casted as a float.
Added unit testing for all the transform functions.

ongtw commented 2 years ago

| Is there a good design (pythonic way) to only check the type once?

I think a bigger question ought to be: is it the intended design to check the type for x once? What if x is changed between the only_once_type_check and subsequent_processing? Or does it make sense to check type on demand, to ensure correctness?