feat(lidar_centerpoint): remove libtorch from centerpoint

[yukke42]

Related Issue(required)

#488

Description(required)

• remove libtorch and implement the equivalent libraries for this package using c++/cuda
• change converted onnx model format, but the pre-trained model is the same as before
• update links to download trained models

before:

• GPU memory: 2817MiB
• inference time: 48.79 ± 3.78 ms

after:

• GPU memory: 1295MiB
• inference time: 45.43 ± 3.37 ms

Review Procedure(required)

Related PR(optional)

Pre-Review Checklist for the PR Author

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

• Read commit-guidelines
• Assign PR to reviewer

If you are adding new package following items are required:

• Documentation with description of the package is available
• A sample launch file and parameter file are available if the package contains executable nodes

Checklist for the PR Reviewer

Reviewers should check the checkboxes below before approval.

• Commits are properly organized and messages are according to the guideline
• 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 / build-and-test-pr: Required to pass before the merge.
• Build and test for PR / clang-tidy-pr: NOT required to pass before the merge. It is up to the reviewer(s). Found false positives? See the [guidelines][clang-tidy-guidelines].
• Check spelling: NOT required to pass before the merge. It is up to the reviewer(s). See here if you want to add some words to the spell check dictionary.

[codecov]

Codecov Report

Merging #480 (88d44c3) into main (ca59048) will decrease coverage by 10.76%.
The diff coverage is 0.00%.

@@            Coverage Diff             @@
##             main    #480       +/-   ##
==========================================
- Coverage   10.76%   0.00%   -10.77%     
==========================================
  Files         712      16      -696     
  Lines       50569     402    -50167     
  Branches     6572       0     -6572     
==========================================
- Hits         5445       0     -5445     
+ Misses      40717     402    -40315     
+ Partials     4407       0     -4407     

Impacted Files	Coverage Δ
.../lidar_centerpoint/lib/include/centerpoint_trt.hpp	0.00% <0.00%> (ø)
...ption/lidar_centerpoint/lib/include/cuda_utils.hpp	0.00% <ø> (ø)
...nterpoint/lib/include/pointcloud_densification.hpp	0.00% <ø> (ø)
...lidar_centerpoint/lib/include/tensorrt_wrapper.hpp	0.00% <ø> (ø)
.../lidar_centerpoint/lib/include/voxel_generator.hpp	0.00% <ø> (ø)
...tion/lidar_centerpoint/lib/src/centerpoint_trt.cpp	0.00% <0.00%> (ø)
...ion/lidar_centerpoint/lib/src/circle_nms_kernel.cu	0.00% <0.00%> (ø)
...rception/lidar_centerpoint/lib/src/network_trt.cpp	0.00% <0.00%> (ø)
...r_centerpoint/lib/src/pointcloud_densification.cpp	0.00% <0.00%> (ø)
...on/lidar_centerpoint/lib/src/postprocess_kernel.cu	0.00% <0.00%> (ø)
... and 729 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ca59048...88d44c3. Read the comment docs.

[yukkysaito]

@yukke42 pls add related PR

[yukke42]

@yukkysaito
libtorch isn't installed in https://github.com/autowarefoundation/autoware

[yukkysaito]

@yukke42 cc @kenji-miyake
how are developpers doing without libtorch?

Does this mean that detection cannot work in https://github.com/autowarefoundation/autoware?

[yukke42]

@.yukke42 cc @.kenji-miyake how are developpers doing without libtorch?

Does this mean that detection cannot work in https://github.com/autowarefoundation/autoware?

@yukkysaito
We used cuda 11.1 and tensorrt 7.2.1, but https://github.com/autowarefoundation/autoware requires cuda 11.4 and tensorrt 8.2.2. (here). Perception packages requiring CUDA aren't supported yet. So, we should make sure that lidar_centerpoint and other packages, s.t. lidar_apollo_instance_segmentation, tensorrt_yolo, ..etc, work on both environment at the same time.

[yukke42]

@yukkysaito @aohsato @mkuri It's now ready for review.

[yukkysaito]

@aohsato @mkuri can you review this PR?

[mkuri]

@miursh ?

[miursh]

I checked it's working in my pc.
before:
GPU memory: 2819MiB
inference time: 31.94 ± 6.92 ms

after:
GPU memory: 1299MiB
inference time: 27.23 ± 6.94 ms

[kaancolak]

I tested it in my computer with sample-map-rosbag and it's working well.

Currently, documentation of rosbag replay simulation not working properly.

[yukke42]

I tested it in my computer with sample-map-rosbag and it's working well.

Currently, documentation of rosbag replay simulation not working properly.

@kaancolak Thank you for testing this PR.
Cloud you create an issue in https://github.com/autowarefoundation/autoware-documentation and describe the problem in more detail?

[yukke42]

build-and-test-differential / build-and-test-differential (pull_request) wasn't passed because the model failed to be downloaded from gdrive.

https://github.com/autowarefoundation/autoware.universe/runs/5640881412?check_suite_focus=true

[yukke42]

build-and-test-differential / build-and-test-differential (pull_request) wasn't passed because the model failed to be downloaded from gdrive.

https://github.com/autowarefoundation/autoware.universe/runs/5640881412?check_suite_focus=true

I will remove to check downloaded files to pass ci.

[miursh]

@wep21 Since I'm not familiar with CUDA programming, could you help me reviewing this PR?

[wep21]

@miursh I have taken a look at the part of cuda implementation, I guess there is no problem.

[wep21]

LGTM