feat(lidar_apollo_segmentation_tvm): port package

[ambroise-arm]

Description

merge #893 first

Port apollo_lidar_segmentation from Autoware.Auto, which uses TVM for the inference. Renamed to match the name of the same package that uses TensorRT.

Related links

Example usage of #628

Tests performed

Ran the unit test locally.

Notes for reviewers

Building

By default the packages don't get built. It requires neural networks from the neural_networks_provider package that only get downloaded with the DOWNLOAD_ARTIFACTS cmake variable set. See its documentation for more information.

Package name

There is already a lidar_apollo_instance_segmentation package that uses TensorRT. I've tried to use the same name with a "_tvm" suffix, but it gets too long and breaks the max line size so I've dropped the "instance" in the name. We may want to rename the current lidar_apollo_instance_segmentation to lidar_apollo_segmentation_tensorrt (or something similar) to align the naming.

Output

The output messages don't match the ones from lidar_apollo_instance_segmentation. It will probably require some follow-up work for them to be interchangeable. But there is no simulator pipeline to test it with at the moment.

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.

[codecov]

Codecov Report

Base: 10.34% // Head: 10.27% // Decreases project coverage by -0.06% ⚠️

Coverage data is based on head (ecd09b8) compared to base (d75ea6e).
Patch has no changes to coverable lines.

❗ Current head ecd09b8 differs from pull request most recent head b7d8a3b. Consider uploading reports for the commit b7d8a3b to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1181      +/-   ##
==========================================
- Coverage   10.34%   10.27%   -0.07%     
==========================================
  Files        1162     1154       -8     
  Lines       82729    82071     -658     
  Branches    19271    19102     -169     
==========================================
- Hits         8555     8436     -119     
+ Misses      64914    64492     -422     
+ Partials     9260     9143     -117     

Flag	Coverage Δ		*Carryforward flag
differential	∅ <ø> (?)
total	10.26% <ø> (-0.07%)	⬇️	Carriedforward from 69758af

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

Impacted Files	Coverage Δ
.../behavior_path_planner/src/turn_signal_decider.cpp	0.00% <0.00%> (-40.97%)	⬇️
...lude/behavior_path_planner/turn_signal_decider.hpp	0.00% <0.00%> (-25.00%)	⬇️
...signal_processing/test/src/lowpass_filter_test.cpp	25.71% <0.00%> (-10.29%)	⬇️
...vehicle_model/sim_model_ideal_steer_acc_geared.cpp	64.15% <0.00%> (-9.44%)	⬇️
...lanner/include/behavior_path_planner/utilities.hpp	25.00% <0.00%> (-1.32%)	⬇️
perception/lidar_centerpoint/src/node.cpp	0.00% <0.00%> (ø)
localization/ekf_localizer/src/measurement.cpp	100.00% <0.00%> (ø)
localization/ndt_scan_matcher/src/util_func.cpp	0.00% <0.00%> (ø)
planning/behavior_velocity_planner/src/node.cpp	0.41% <0.00%> (ø)
localization/ekf_localizer/src/ekf_localizer.cpp	0.00% <0.00%> (ø)
... and 42 more

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.

[ambroise-arm]

Thanks for the review! I am working on addressing the comments.

[ambroise-arm]

The first force-push is for rebasing, the second force-push is the actual change.
I've tested the new output with the rosbag demo here. The quality of the detection is not quite as good as the TensorRT package, but I believe that can be addressed later.

[angry-crab]

Thank you for the update! I tried to run a rosbag test today but my environment was somehow broken and I'm still trying to fix it. Since you were able to run the test, this pr is good to go. As for quality of detection, I think we could open up another issue for that.

[angry-crab]

LGTM

[ambroise-arm]

I have created an issue to track the difference in detection quality.
Can you merge the PR? I don't have write access.