feat(yolox): updating autoware tensorrt yolox for type adaptation

[Owen-Liuyuxuan]

Description

Updating type adaptation for Autoware YOLOX.

Included changes:

1. Adding additional type adaptation packages defining image container and compressed image container.
   • Please check the readme of the package for more information.
2. Adopting the core YOLOX interface to work with type adaptation.
3. Slightly refactoring YOLOX node,
   • Organize part of codes for reusability.
   • Adding options for behavior optimization
     • choosing between activating type adaptation or not.
     • choosing to use image transport (which will process the image on the CPU) or not.
     • choosing to output the debug/rois/image for not (which will move the image to CPU for openCV drawing).

Limitation:

• I have only covered functions that will be used by the tensorrt_yolox_node and functions including "multiScaleFeedForward" are not fully adapted. But it should not be difficult considering the most tricky parts only lie in the function "preprocessGPU"
• Type adaptation now only works with GPU preprocessing.
• We need more considerations on how to work with LiDAR.

Related links

Parent Issue:

• None for the moment.

Private Links:

• TIERIV internal link

How was this PR tested?

For this PR, I launch six Tensorrt YOLOX nodes in parallel and use my personal codes to publish six streams of normal images there.

This test simply turns on/off type adaptation on the launch file, which means that this shows how running type adaptation codes without pipeline optimization does not make performance worse.

Type adaptation ON vs Type adaption OFF:

Tests that include streamlining multiple nodes with type adaptation and getting optimized results are presented in the TIERIV internal link

Notes for reviewers

None.

Interface changes

ROS Parameter Changes

Additions and removals

Change type	Parameter Name	Type	Default Value	Description
Added	is_using_image_transport	bool	true	If true, will use image_transport, false -> will only publish raw images
Added	type_adaptation_activated	bool	true	If true, will use type_adaptation to transfer images in GPU.
Added	is_publish_debug_rois_image_	bool	true	If true, will draw ROIs in the image and publish it for debugging

Effects on system behavior

None.

[github-actions]

Thank you for contributing to the Autoware project!

🚧 If your pull request is in progress, switch it to draft mode.

Please ensure:

• You've checked our contribution guidelines.
• Your PR follows our pull request guidelines.
• All required CI checks pass before marking the PR ready for review.

[ktro2828]

@Owen-Liuyuxuan Thank you for PR!
First of all, as of this conversation, CUDA related packages have been moved under sensing directory.
Then, it looks better to move autoware_type_adapters under sensing or autoware_cuda_utils.

[Owen-Liuyuxuan]

@ktro2828
Thanks for the information. I shifted the entire adapters into the sensing directory.