[amadeuszsz]

Description

This PR refines the PTv3 runtime point cloud handling and output reconstruction behavior. It adds configurable source reconstruction modes so PTv3 can optionally publish predictions back on the original input point layout instead of only voxel representatives. Changes:

• Added source_reconstruction node parameter with modes:
  • none: existing behavior, publish voxelized output only.
  • partial: unvoxelize predictions back to cropped in-range input points.
  • full: unvoxelize and uncrop predictions back to the full input cloud.
    • Out-of-range or clipped points receive unknown label 255 and 0.0 class probabilities.
• Updated segmentation output to include normalized Shannon entropy for uncertainty inspection.
• Updated visualization and filtered point cloud generation to work correctly with reconstructed outputs.
• Added schema/config support for the new reconstruction mode.
• Extra:
  • Remap blackboard's (cuda) topics with dummy suffix. This let's topics negotiation fail explicitly which is expected outcome for nodes in separate process.
  • Fixes an out-of-bounds initialization of hash index buffers by matching the Thrust sequence range to the allocated cloud capacity.

Related links

Parent Issue:

• Link

How was this PR tested?

Average latency in milliseconds for sample rosbag input with RTX 3060 laptop.

Metric	Original	Mode none	Mode partial	Mode full
Preprocessing	1.23	1.43	2.92	2.56
Inference	31.11	32.55	29.63	31.01
Postprocessing	1.19	1.38	3.88	4.05
Total	33.55	35.38	36.44	37.63

Original implementation (before this PR) and mode none seem to have equivalent latency. partial and full come with expected overhead.

I would suggest to keep mode full as default, because intuitively we expect same output points sequence as input.

Notes for reviewers

None.

Interface changes

None.

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.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 55caa03d08

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[codecov]

Codecov Report

❌ Patch coverage is 0% with 110 lines in your changes missing coverage. Please review.
✅ Project coverage is 0.00%. Comparing base (4d6292d) to head (efc7ab1).

Files with missing lines	Patch %	Lines
perception/autoware_ptv3/lib/ptv3_trt.cpp	0.00%	62 Missing ⚠️
.../autoware_ptv3/lib/preprocess/preprocess_kernel.cu	0.00%	22 Missing ⚠️
...utoware_ptv3/lib/postprocess/postprocess_kernel.cu	0.00%	18 Missing ⚠️
...utoware_ptv3/include/autoware/ptv3/ptv3_config.hpp	0.00%	6 Missing ⚠️
perception/autoware_ptv3/src/ptv3_node.cpp	0.00%	2 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##             main   #12547       +/-   ##
===========================================
- Coverage   18.64%    0.00%   -18.65%     
===========================================
  Files        1918       11     -1907     
  Lines      131362      761   -130601     
  Branches    44502        0    -44502     
===========================================
- Hits        24486        0    -24486     
+ Misses      86760      761    -85999     
+ Partials    20116        0    -20116     

Flag	Coverage Δ
full-suite	0.00% <0.00%> (-18.65%)	⬇️

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

This PR extends the autoware_ptv3 segmentation pipeline to optionally reconstruct PTv3 voxel predictions back onto the original (cropped or full) source point layout, and augments the segmented output with a per-point normalized Shannon entropy field for uncertainty inspection.

Changes:

• Added source_reconstruction parameter (none/partial/full) and implemented partial/full reconstruction paths (inverse mapping + reconstruction kernels).
• Extended segmented pointcloud output with an entropy field and updated visualization/filtered outputs to operate on reconstructed outputs.
• Updated schema/config/launch to support the new parameter and adjust CUDA topic remappings.

Reviewed changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
perception/autoware_ptv3/src/ptv3_node.cpp	Declares source_reconstruction ROS parameter and forwards it into PTv3Config.
perception/autoware_ptv3/schema/ptv3.schema.json	Adds schema support for source_reconstruction (enum/default/required).
perception/autoware_ptv3/lib/ptv3_trt.cpp	Allocates reconstruction buffers, selects output capacity based on reconstruction mode, clears buffers, and switches postprocess outputs to reconstructed sources.
perception/autoware_ptv3/lib/preprocess/preprocess_kernel.cu	Adds inverse-map scatter kernel and expands feature generation to optionally emit source features/cropped source points/inverse mapping.
perception/autoware_ptv3/lib/postprocess/postprocess_kernel.cu	Adds entropy to segmentation output and introduces partial/full reconstruction CUDA kernels + safer visualization color lookup.
perception/autoware_ptv3/launch/ptv3.launch.xml	Remaps CUDA topics to .../cuda/dummy to force explicit negotiation failures across processes.
perception/autoware_ptv3/include/autoware/ptv3/ptv3_trt.hpp	Adds state for source/cropped counts, raw input pointer, and reconstruction buffers.
perception/autoware_ptv3/include/autoware/ptv3/ptv3_config.hpp	Introduces SourceReconstruction enum + parsing, and threads the new config through the constructor.
perception/autoware_ptv3/include/autoware/ptv3/preprocess/preprocess_kernel.hpp	Updates generateFeatures() signature and exposes crop mask/indices for full reconstruction.
perception/autoware_ptv3/include/autoware/ptv3/postprocess/postprocess_kernel.hpp	Updates visualization API and adds reconstruction method declarations.
perception/autoware_ptv3/config/ptv3.param.yaml	Sets a default source_reconstruction mode in the shipped param file.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[mojomex]

Temporary changes made it into the PR

[manato]

@amadeuszsz
Thank you for your contribution!
I left a few comments on the CUDA code. Your consideration would be appreciated!

[manato]

Since my concerns have been treated, and it seems there are no remaining untreated comments from other reviews, I approve this PR.
Thank you for your contribution!