feat(lidar_centerpoint): accelerated preprocessing for centerpoint

[knzo25]

Description

While changing the point type to the new standard, realized that preprocessing was bottleneck in centerpoint.
Apologies to Amadeusz since he did almost the same already in transfusion, and I forgot until writing this PR but I also moved the preprocessing to cuda for gpu acceleration.

During tests in my desktop the detection time went from 16.9ms to 13.0ms, making preprocessing about ~1ms if we forcefully add syncs to measure the time.

Related links

Tests performed

• Tested using data from the taxi

Notes for reviewers

Interface changes

ROS Topic Changes

ROS Parameter Changes

Effects on system behavior

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.
• The PR has been properly tested.
• The PR has been reviewed by the code owners.

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.
• The PR is ready for merge.

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

[knzo25]

@kminoda
This is by no means urgent, but it is always nice to reduce the perception latency.

if you do not feel comfortable reviewing this PR, Uetake-san, Hirabayashi-san, Tanaka-san, and Amadeusz are good candidates)

[kminoda]

@knzo25 Thank you! And yes, I am still catching up with this package (especially the CUDA part), so would be nice if someone else could review it as a co-reviewer 🙏

@amadeuszsz Hi, would you help me out reviewing CUDA related code?

[kminoda]

@knzo25 And also, would you execute Autoware Evaluator and verify that the basic scenarios would pass? Let me know if you don't know how to execute it.

[kminoda]

Thank you for the awesome PR 🚀

[knzo25]

@kminoda
Regarding the evaluation I will look into it as soon as I can (got other stuff to deliver first)

[amadeuszsz]

@amadeuszsz Hi, would you help me out reviewing CUDA related code?

@kminoda
Sure! I would like to make a review next week. Please let me know if it's urgent, I might change my priorities.
I guess you can't add me as a reviewer yet. I added my account to the contributors sheet, so I hope we will be able assign me soon.

[kminoda]

@amadeuszsz Thank you. Not an urgent task, so please review this when you have time.

[amadeuszsz]

Highly performance improvement observed! 👍🏻
Apart of minor changes requested, please check issues on your desktop with large clouds size & pointer arithmetic which causes wrong detections.

[knzo25]

Note to self: Pointpainting was using some of centerpoint parts as a base clase, so I will have to port some of the changes to pointpainting or duplicate logic for pointpainting

[knzo25]

@kminoda @YoshiRi @tzhong518
Since pointainting inherited some logic from centerpoint, and centerpoint now does things differently for GPU acceleration, pointpainting broke. For now, I just copied the logic from centerpoint before this PR and added it to the pointpainting package (the same optimization could not be applied easily).

If you also want pointpainting accelerated, let's us discuss / schedule it a later date please

[kminoda]

@knzo25 Thanks. Still haven't understood in depth, but just to confirm: we would anyway have to implement different logic for centerpoint and pointpainting, right? I vote for merging this PR independently from pointpainting's further acceleration 👍

[amadeuszsz]

From my side, lidar_centerpoint is ready to merge! 🚀 Please, confirm if the pointpainting discussion will be solved and we can proceed to merge.

[knzo25]

@kminoda
I will wait until the message migration happens before attempting to merge this. In addition I want to learn to use the evaluator so I will be trying that as soon as I get a respite from other things 🙏 (edit: now it is properly blocked xD)

[kminoda]

@knzo25 Hi, would you fix the cppcheck error? (see cppcheck-differential result)

perception/lidar_centerpoint/include/lidar_centerpoint/network/tensorrt_wrapper.hpp:34:4: warning: inconclusive: Class 'TensorRTWrapper' which has virtual members does not have a virtual destructor. [virtualDestructor]
~TensorRTWrapper();

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 3.21%. Comparing base (507e3f4) to head (0a8986f).
Report is 65 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #6989       +/-   ##
==========================================
- Coverage   14.84%   3.21%   -11.63%     
==========================================
  Files        1999     126     -1873     
  Lines      139163    8294   -130869     
  Branches    43716    1402    -42314     
==========================================
- Hits        20661     267    -20394     
+ Misses      95731    7971    -87760     
+ Partials    22771      56    -22715     

Flag	Coverage Δ
differential	3.21% <ø> (?)
total	?

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[xmfcx]

@knzo25 starting from this PR, The CI for build-main started failing:

• https://github.com/autowarefoundation/autoware.universe/actions/runs/9590002364/job/26444678737#step:15:22008

8: C++ exception with description "cudaErrorInsufficientDriver (35)@/__w/autoware.universe/autoware.universe/perception/lidar_centerpoint/include/lidar_centerpoint/cuda_utils.hpp#L80: CUDA driver version is insufficient for CUDA runtime version" thrown in the test body.

For some reason the has CI passed for the build-and-test-differential cuda yet it fails the build-and-test checks.