feat(tensorrt_yolox): add build only option for tensorrt engine

[yukke42]

Description

• the option to build tensorrt engine without inference is required for test on cloud
• add depend package
• fix documentation

Pre-review checklist for the PR author

The PR author must check the checkboxes below when creating the PR.

• I've confirmed the contribution guidelines.
• The PR follows the pull request guidelines.

In-review checklist for the PR reviewers

The PR reviewers must check the checkboxes below before approval.

• The PR follows the pull request guidelines.

Post-review checklist for the PR author

The PR author must check the checkboxes below before merging.

• There are no open discussions or they are tracked via tickets.

After all checkboxes are checked, anyone who has write access can merge the PR.

[codecov]

Codecov Report

Patch coverage has no change and project coverage change: -0.05 ⚠️

Comparison is base (f9075cb) 12.81% compared to head (5ee5293) 12.76%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2982      +/-   ##
==========================================
- Coverage   12.81%   12.76%   -0.05%     
==========================================
  Files        1212     1216       +4     
  Lines       85613    85922     +309     
  Branches    24260    24260              
==========================================
  Hits        10968    10968              
- Misses      63282    63591     +309     
  Partials    11363    11363              

Flag	Coverage Δ		*Carryforward flag
differential	0.00% <0.00%> (?)
total	12.81% <ø> (ø)		Carriedforward from 4536ca8

*This pull request uses carry forward flags. Click here to find out more.

Impacted Files	Coverage Δ
...ception/tensorrt_yolox/src/tensorrt_yolox_node.cpp	0.00% <0.00%> (ø)
...havior_path_planner/behavior_path_planner_node.hpp	0.00% <ø> (ø)
...th_planner/scene_module/scene_module_interface.hpp	0.00% <ø> (ø)
...or_path_planner/src/behavior_path_planner_node.cpp	0.15% <ø> (ø)
...rt_yolox/src/yolox_single_image_inference_node.cpp	0.00% <0.00%> (ø)
perception/tensorrt_yolox/src/tensorrt_yolox.cpp	0.00% <0.00%> (ø)
...rt_yolox/include/tensorrt_yolox/tensorrt_yolox.hpp	0.00% <0.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[wep21]

@yukke42 Why don't you use trtexec?

[yukke42]

@yukke42 Why don't you use trtexec?

@wep21
If the tensorrt_yolox package becomes more complicated in terms of cuda/tensorrt options, s.t. int8, calibration, ... etc,
trtexec can be used by develpers who knows both trtexec and this package. But, by this option, they don't need know them and just add the build_only arg to launch command.

[wep21]

they don't need know them and just add the build_only arg to launch command.

@yukke42
But they do need to know launch options such as int8, calibrator defined in each package(not only this package but other packages which use tensorrt), right?
I think trtexec is a generalized tool which covers most use cases to build tensorrt engine file, so we don't have to add build only option in each package.

[yukke42]

But they do need to know launch options such as int8, calibrator defined in each package(not only this package but other packages which use tensorrt), right?
I think trtexec is a generalized tool which covers most use cases to build tensorrt engine file, so we don't have to add build only option in each package.

@wep21 Yes, we can do the same thing with this option and trtexec. But it's hard to guarantee that the same engine file is generated as we expected.
tensorrt_yolox uses optimizationProfile and I think it's --optShapes of trtexec. It requires the magic number like this and another scirpt to get values from onnx file to delete these magic numbers. And, traffic_light_classifier don't use optimizationProfile , the package don't need opt option.
If some contrbutes change the tensorrt option/implementation, we need to catch up all the chages and change trtexec options as well. This is for not only this package but also other packages which require tensorrt. And how do we support trtexec options for many autoware versions?

[wep21]

It may be out of scope from this PR, but I think the problem is that each package has own implementation to build tensorrt engine, while they do almost the same process. I think it is better to use trtexec to build tensorrt engine in launch file like

<launch>
  <executable cmd="trtexec --onnx=$(var onnx) --saveEngine=$(var enigne) --optShapes=input:1x3x$(var input_h)x$(var input_w)" name="trt_build" output="screen" shell="true"/>

  <node pkg="yolo" exec="yolo" name="yolo">
    <param name="engine" value="$(var engine)" />
  </node>
<launch>

and then remove engine build function from each package. What do you think?
Anyway, I won't block this PR if you need this feature as soon as possible for cloud test.

[yukke42]

@wep21 I see, Let me make a new discussion topic about it.
I hope you allow this pr to merge and then I'll chnage it later.

[wep21]

LGTM