Is your feature request related to a problem? Please describe.
I have been using the updated 3.0.0rc13 version for inference with pytorch and thankfully get similar performance as tensorflow implementations (80-120it/s). This is done in a custom conda environment, which I happened to nuke recently, so I sought out docker for a quicker implementation fix. Unfortunately, when i go to use the official, 'latest' docker images, they utilize 3.0.0rc8 which shows catastrophic performance (1-3 it/s) when using the pytorch models.

Describe the solution you'd like
I understand that making a new docker image for every minute version change is not practical but given the performance increases, it'd be nice to have 3.0.0rc13 available, even if just in a beta version and not the official 'latest' image.

Thank you all so much for the work you do!