feat(hesai): implement mask-based pruning filter for Hesai

[mojomex]

PR Type

• New Feature

Related Links

• feat: mask-based pruning filter (vendor independent) #250 -- based on this PR

Description

This PR implements the mask-based pruning filter introduced in #250 for all Hesai sensors. For more specifics on sensor configuration changes and filter behavior, see #250.

Review Procedure

• Test whether pointcloud output is as expected when not specifying a filter
• Try and create a downsample mask, e.g. in GIMP, and see if it works expected

Pre-Review Checklist for the PR Author

PR Author should check the checkboxes below when creating the PR.

• Assign PR to reviewer

Checklist for the PR Reviewer

Reviewers should check the checkboxes below before approval.

• Commits are properly organized and messages are according to the guideline
• (Optional) Unit tests have been written for new behavior
• PR title describes the changes

Post-Review Checklist for the PR Author

PR Author should check the checkboxes below before merging.

• All open points are addressed and tracked via issues or tickets

CI Checks

• Build and test for PR: Required to pass before the merge.

[mojomex]

🟢 Self-Evaluation

🟢 Performance

Take-aways:

• Performance impact is near 0 or negative (as in, performance improves when using the filter)
  • A disabled filter, and a filter that keeps all points, perform as well as the baseline before this PR
  • A filter discarding 50 % of points improves mean decode time by 7.4 %
  • A filter discarding all points improves mean decode time by 30.3 %

The below performance measurements have been performed with a modified decoder that measures timings of the convert_returns function and accumulates them for one scan. The filter for invalid points (distance == 0) has been disabled, as have other distance checks, to keep the number of input points artificially high ("worst-case"). The decoded number of points per cloud was around 460.000 which is around 2/3 of the maximum possible output in high-res+dual-return mode for OT128.

label	('total_time', 'count')	('total_time', 'mean')	('total_time', 'std')	('total_time', 'min')	('total_time', '50%')	('total_time', 'max')
baseline	942	10.3296	0.390481	9.06057	10.3843	12.149
disabled	941	10.1409	0.296578	9.07079	10.0777	10.852
discard_all	941	7.19801	0.175469	6.94406	7.22401	7.68576
discard_half	940	9.56162	0.22284	9.16713	9.4687	10.6639
keep_all	940	10.2512	0.278572	9.13494	10.172	10.8176

[codecov]

Codecov Report

Attention: Patch coverage is 37.77778% with 28 lines in your changes missing coverage. Please review.

Project coverage is 29.28%. Comparing base (9e40e2e) to head (2e8929b).
Report is 7 commits behind head on main.

Files with missing lines	Patch %	Lines
nebula_ros/src/hesai/hesai_ros_wrapper.cpp	28.00%	15 Missing and 3 partials ⚠️
...s/nebula_decoders_hesai/decoders/hesai_decoder.hpp	44.44%	3 Missing and 2 partials ⚠️
...ommon/include/nebula_common/hesai/hesai_common.hpp	33.33%	2 Missing ⚠️
nebula_ros/src/hesai/decoder_wrapper.cpp	50.00%	2 Missing ⚠️
...ecoders/src/nebula_decoders_hesai/hesai_driver.cpp	66.66%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #251      +/-   ##
==========================================
- Coverage   29.32%   29.28%   -0.04%     
==========================================
  Files         108      112       +4     
  Lines        9580     9733     +153     
  Branches     3177     3252      +75     
==========================================
+ Hits         2809     2850      +41     
- Misses       6265     6359      +94     
- Partials      506      524      +18     

Flag	Coverage Δ
differential	29.28% <37.77%> (?)
total	?

Flags with carried forward coverage won't be shown. Click here to find out more.

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

[mojomex]

@drwnz Rebased onto the updated #250, added missing support for XT16, please check 🙏

[mojomex]

@drwnz I changed the std::string config item to std::optional<std::string> as discussed offline. I also added clearer instructions on which color settings to use to produce correct grayscale images.

[drwnz]

Awesome, LGTM!