feat(centerpoint): integrate CenterPoint into unified deployment pipeline by vividf · Pull Request #161 · tier4/AWML

vividf · 2025-12-25T08:30:38Z

Summary

Adds a CenterPoint project bundle under deployment/projects/centerpoint/ and hooks it into the unified deployment CLI via the project registry.
Supports PyTorch / ONNXRuntime / TensorRT export + evaluation with a componentized (multi-file) ONNX → per-component TRT engine flow.
Removes legacy CenterPoint deployment entrypoints in projects/CenterPoint/* in favor of the unified framework.

Change points

New CenterPoint deployment project
Project registration + CLI flag: deployment/projects/centerpoint/cli.py (--rot-y-axis-reference)
Wiring/entrypoint: deployment/projects/centerpoint/entrypoint.py
Runtime config: deployment/projects/centerpoint/config/deploy_config.py
Dataloader / evaluator / model building:
Backend pipelines (staged inference aligned across backends):
Export pipelines
- ONNX multi-file export using component extraction:
  - deployment/projects/centerpoint/export/component_extractor.py
  - deployment/projects/centerpoint/export/onnx_export_pipeline.py
- TensorRT engine-per-onnx conversion:
  - deployment/projects/centerpoint/export/tensorrt_export_pipeline.py
ONNX model variants moved under deployment:
deployment/projects/centerpoint/onnx_models
Docs / packaging
- Adds projects/CenterPoint/Dockerfile for deployment dependencies (onnxruntime-gpu, onnxsim, pycuda, tensorrt-cu12).
- Updates projects/CenterPoint/README.md with unified deployment usage example.

How to run

python -m deployment.cli.main centerpoint   deployment/projects/centerpoint/config/deploy_config.py   projects/CenterPoint/configs/t4dataset/Centerpoint/second_secfpn_4xb16_121m_j6gen2_base_t4metric_v2.py   --rot-y-axis-reference

Exported ONNX (Same)

Voxel Encoder

Backbone Head

Evaluation result (Same for Deployment and Test)

Test with test.py

python tools/detection3d/test.py projects/CenterPoint/configs/t4dataset/Centerpoint/second_secfpn_4xb16_121m_j6gen2_base_t4metric_v2.py work_dirs/centerpoint/best_checkpoint.pth

Frame:
Total Num: 19
Skipped Frames: []
Skipped Frames Count: 0

Ground Truth Num: 650

mAP: 0.5860, mAPH: 0.5621 (Center Distance BEV)

Label: car

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	583	337	0.7919	0.7744	0.8207	0.9007	0.7537	0.331500
1.00	583	337	0.8425	0.8242	0.8507	0.9266	0.7864	0.321666
2.00	583	337	0.8607	0.8421	0.8957	0.9755	0.8279	0.321666
4.00	583	337	0.8607	0.8421	0.8957	0.9755	0.8279	0.321666

Label: truck

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	278	152	0.5563	0.5457	0.7237	0.8857	0.6118	0.358953
1.00	278	152	0.8317	0.8122	0.8671	0.9254	0.8158	0.225477
2.00	278	152	0.8459	0.8276	0.8741	0.9328	0.8224	0.225477
4.00	278	152	0.8675	0.8476	0.8741	0.9328	0.8224	0.225477

Label: bus

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849640
1.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849640
2.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849640
4.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849640

Label: bicycle

Threshold	Predict_num	Groundtruth_num	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	1	19	nan	nan	nan	nan
1.00	1	19	nan	nan	nan	nan
2.00	1	19	nan	nan	nan	nan
4.00	1	19	nan	nan	nan	nan

Label: pedestrian

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	328	123	0.3030	0.2310	0.5123	0.4506	0.5935	0.222946
1.00	328	123	0.3030	0.2310	0.5123	0.4506	0.5935	0.222946
2.00	328	123	0.3166	0.2417	0.5193	0.4568	0.6016	0.222946
4.00	328	123	0.3489	0.2668	0.5422	0.4306	0.7317	0.180424

Summary:

Label	Predict_num	GT_nums	Thresholds	mean AP	APs	Mean APH	APHs
car	583	337	0.50/1.00/2.00/4.00	0.8389	0.7919 / 0.8425 / 0.8607 / 0.8607	0.821	0.7744 / 0.8242 / 0.8421 / 0.8421
truck	278	152	0.50/1.00/2.00/4.00	0.7754	0.5563 / 0.8317 / 0.8459 / 0.8675	0.758	0.5457 / 0.8122 / 0.8276 / 0.8476
bus	23	19	0.50/1.00/2.00/4.00	0.9979	0.9979 / 0.9979 / 0.9979 / 0.9979	0.989	0.9889 / 0.9889 / 0.9889 / 0.9889
bicycle	1	19	0.50/1.00/2.00/4.00	0.0000	0.0000 / 0.0000 / 0.0000 / 0.0000	0.000	0.0000 / 0.0000 / 0.0000 / 0.0000
pedestrian	328	123	0.50/1.00/2.00/4.00	0.3179	0.3030 / 0.3030 / 0.3166 / 0.3489	0.243	0.2310 / 0.2310 / 0.2417 / 0.2668

mAP: 0.6076, mAPH: 0.5831 (Plane Distance)

Label: car

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	583	337	0.8602	0.8422	0.8957	0.9755	0.8279	0.321666
4.00	583	337	0.8607	0.8421	0.8957	0.9755	0.8279	0.321666

Label: truck

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	278	152	0.8410	0.8281	0.8671	0.9254	0.8158	0.225477
4.00	278	152	0.8526	0.8329	0.8741	0.9328	0.8224	0.225477

Label: bus

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849640
4.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849640

Label: bicycle

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	1	19	0.0000	0.0000	nan	nan	nan	nan
4.00	1	19	0.0000	0.0000	nan	nan	nan	nan

Label: pedestrian

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	328	123	0.3166	0.2417	0.5193	0.4568	0.6016	0.222946
4.00	328	123	0.3490	0.2664	0.5422	0.4306	0.7317	0.180424

Summary:

Label	Predict_num	GT_nums	Thresholds	mean AP	APs	Mean APH	APHs
car	583	337	2.00/4.00	0.8604	0.8602 / 0.8607	0.842	0.8422 / 0.8421
truck	278	152	2.00/4.00	0.8468	0.8410 / 0.8526	0.831	0.8281 / 0.8329
bus	23	19	2.00/4.00	0.9979	0.9979 / 0.9979	0.989	0.9889 / 0.9889
bicycle	1	19	2.00/4.00	0.0000	0.0000 / 0.0000	0.000	0.0000 / 0.0000
pedestrian	328	123	2.00/4.00	0.3328	0.3166 / 0.3490	0.254	0.2417 / 0.2664

Test with Deployment pipeline

python -m deployment.cli.main centerpoint   deployment/projects/centerpoint/config/deploy_config.py   projects/CenterPoint/configs/t4dataset/Centerpoint/second_secfpn_4xb16_121m_j6gen2_base_t4metric_v2.py   --rot-y-axis-reference

PYTORCH Results:

mAP: 0.5860, mAPH: 0.5621 (Center Distance BEV)

Label: car

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	583	337	0.7919	0.7744	0.8207	0.9007	0.7537	0.331500
1.00	583	337	0.8425	0.8242	0.8507	0.9266	0.7864	0.321666
2.00	583	337	0.8607	0.8421	0.8957	0.9755	0.8279	0.321666
4.00	583	337	0.8607	0.8421	0.8957	0.9755	0.8279	0.321666

Label: truck

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	278	152	0.5563	0.5457	0.7237	0.8857	0.6118	0.358953
1.00	278	152	0.8317	0.8122	0.8671	0.9254	0.8158	0.225477
2.00	278	152	0.8459	0.8276	0.8741	0.9328	0.8224	0.225477
4.00	278	152	0.8675	0.8476	0.8741	0.9328	0.8224	0.225477

Label: bus

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849640
1.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849640
2.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849640
4.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849640

Label: bicycle

Threshold	Predict_num	Groundtruth_num	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	1	19	nan	nan	nan	nan
1.00	1	19	nan	nan	nan	nan
2.00	1	19	nan	nan	nan	nan
4.00	1	19	nan	nan	nan	nan

Label: pedestrian

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	328	123	0.3030	0.2310	0.5123	0.4506	0.5935	0.222946
1.00	328	123	0.3030	0.2310	0.5123	0.4506	0.5935	0.222946
2.00	328	123	0.3166	0.2417	0.5193	0.4568	0.6016	0.222946
4.00	328	123	0.3489	0.2668	0.5422	0.4306	0.7317	0.180424

Summary:

Label	Predict_num	GT_nums	Thresholds	mean AP	APs	Mean APH	APHs
car	583	337	0.50/1.00/2.00/4.00	0.8389	0.7919 / 0.8425 / 0.8607 / 0.8607	0.821	0.7744 / 0.8242 / 0.8421 / 0.8421
truck	278	152	0.50/1.00/2.00/4.00	0.7754	0.5563 / 0.8317 / 0.8459 / 0.8675	0.758	0.5457 / 0.8122 / 0.8276 / 0.8476
bus	23	19	0.50/1.00/2.00/4.00	0.9979	0.9979 / 0.9979 / 0.9979 / 0.9979	0.989	0.9889 / 0.9889 / 0.9889 / 0.9889
bicycle	1	19	0.50/1.00/2.00/4.00	0.0000	0.0000 / 0.0000 / 0.0000 / 0.0000	0.000	0.0000 / 0.0000 / 0.0000 / 0.0000
pedestrian	328	123	0.50/1.00/2.00/4.00	0.3179	0.3030 / 0.3030 / 0.3166 / 0.3489	0.243	0.2310 / 0.2310 / 0.2417 / 0.2668

mAP: 0.6076, mAPH: 0.5831 (Plane Distance)

Label: car

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	583	337	0.8602	0.8422	0.8957	0.9755	0.8279	0.321666
4.00	583	337	0.8607	0.8421	0.8957	0.9755	0.8279	0.321666

Label: truck

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	278	152	0.8410	0.8281	0.8671	0.9254	0.8158	0.225477
4.00	278	152	0.8526	0.8329	0.8741	0.9328	0.8224	0.225477

Label: bus

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849640
4.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849640

Label: bicycle

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	1	19	0.0000	0.0000	nan	nan	nan	nan
4.00	1	19	0.0000	0.0000	nan	nan	nan	nan

Label: pedestrian

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	328	123	0.3166	0.2417	0.5193	0.4568	0.6016	0.222946
4.00	328	123	0.3490	0.2664	0.5422	0.4306	0.7317	0.180424

Summary:

Label	Predict_num	GT_nums	Thresholds	mean AP	APs	Mean APH	APHs
car	583	337	2.00/4.00	0.8604	0.8602 / 0.8607	0.842	0.8422 / 0.8421
truck	278	152	2.00/4.00	0.8468	0.8410 / 0.8526	0.831	0.8281 / 0.8329
bus	23	19	2.00/4.00	0.9979	0.9979 / 0.9979	0.989	0.9889 / 0.9889
bicycle	1	19	2.00/4.00	0.0000	0.0000 / 0.0000	0.000	0.0000 / 0.0000
pedestrian	328	123	2.00/4.00	0.3328	0.3166 / 0.3490	0.254	0.2417 / 0.2664

ONNX Results:

mAP: 0.5860, mAPH: 0.5619 (Center Distance BEV)

Label: car

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	580	337	0.7919	0.7745	0.8207	0.9007	0.7537	0.332124
1.00	580	337	0.8425	0.8242	0.8507	0.9266	0.7864	0.325795
2.00	580	337	0.8606	0.8414	0.8957	0.9755	0.8279	0.325795
4.00	580	337	0.8606	0.8414	0.8957	0.9755	0.8279	0.325795

Label: truck

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	278	152	0.5516	0.5411	0.7237	0.8857	0.6118	0.358248
1.00	278	152	0.8307	0.8106	0.8662	0.9318	0.8092	0.227877
2.00	278	152	0.8454	0.8265	0.8732	0.9394	0.8158	0.227877
4.00	278	152	0.8672	0.8466	0.8732	0.9394	0.8158	0.227877

Label: bus

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849546
1.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849546
2.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849546
4.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849546

Label: bicycle

Threshold	Predict_num	Groundtruth_num	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	1	19	nan	nan	nan	nan
1.00	1	19	nan	nan	nan	nan
2.00	1	19	nan	nan	nan	nan
4.00	1	19	nan	nan	nan	nan

Label: pedestrian

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	330	123	0.3050	0.2325	0.5123	0.4506	0.5935	0.221067
1.00	330	123	0.3050	0.2325	0.5123	0.4506	0.5935	0.221067
2.00	330	123	0.3185	0.2431	0.5193	0.4568	0.6016	0.221067
4.00	330	123	0.3506	0.2683	0.5410	0.4320	0.7236	0.183189

Summary:

Label	Predict_num	GT_nums	Thresholds	mean AP	APs	Mean APH	APHs
car	580	337	0.50/1.00/2.00/4.00	0.8389	0.7919 / 0.8425 / 0.8606 / 0.8606	0.820	0.7745 / 0.8242 / 0.8414 / 0.8414
truck	278	152	0.50/1.00/2.00/4.00	0.7737	0.5516 / 0.8307 / 0.8454 / 0.8672	0.756	0.5411 / 0.8106 / 0.8265 / 0.8466
bus	23	19	0.50/1.00/2.00/4.00	0.9979	0.9979 / 0.9979 / 0.9979 / 0.9979	0.989	0.9889 / 0.9889 / 0.9889 / 0.9889
bicycle	1	19	0.50/1.00/2.00/4.00	0.0000	0.0000 / 0.0000 / 0.0000 / 0.0000	0.000	0.0000 / 0.0000 / 0.0000 / 0.0000
pedestrian	330	123	0.50/1.00/2.00/4.00	0.3198	0.3050 / 0.3050 / 0.3185 / 0.3506	0.244	0.2325 / 0.2325 / 0.2431 / 0.2683

mAP: 0.6072, mAPH: 0.5830 (Plane Distance)

Label: car

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	580	337	0.8541	0.8409	0.8957	0.9755	0.8279	0.325795
4.00	580	337	0.8606	0.8414	0.8957	0.9755	0.8279	0.325795

Label: truck

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	278	152	0.8402	0.8278	0.8662	0.9318	0.8092	0.227877
4.00	278	152	0.8523	0.8319	0.8732	0.9394	0.8158	0.227877

Label: bus

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849546
4.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849546

Label: bicycle

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	1	19	0.0000	0.0000	nan	nan	nan	nan
4.00	1	19	0.0000	0.0000	nan	nan	nan	nan

Label: pedestrian

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	330	123	0.3185	0.2431	0.5193	0.4568	0.6016	0.221067
4.00	330	123	0.3506	0.2676	0.5433	0.4292	0.7398	0.179694

Summary:

Label	Predict_num	GT_nums	Thresholds	mean AP	APs	Mean APH	APHs
car	580	337	2.00/4.00	0.8574	0.8541 / 0.8606	0.841	0.8409 / 0.8414
truck	278	152	2.00/4.00	0.8463	0.8402 / 0.8523	0.830	0.8278 / 0.8319
bus	23	19	2.00/4.00	0.9979	0.9979 / 0.9979	0.989	0.9889 / 0.9889
bicycle	1	19	2.00/4.00	0.0000	0.0000 / 0.0000	0.000	0.0000 / 0.0000
pedestrian	330	123	2.00/4.00	0.3345	0.3185 / 0.3506	0.255	0.2431 / 0.2676

TENSORRT Results:

mAP: 0.5866, mAPH: 0.5629 (Center Distance BEV)

Label: car

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	586	337	0.7896	0.7718	0.8148	0.8908	0.7507	0.330931
1.00	586	337	0.8421	0.8235	0.8480	0.9201	0.7864	0.323492
2.00	586	337	0.8604	0.8412	0.8928	0.9688	0.8279	0.323492
4.00	586	337	0.8604	0.8412	0.8928	0.9688	0.8279	0.323492

Label: truck

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	282	152	0.5480	0.5375	0.7160	0.8762	0.6053	0.361066
1.00	282	152	0.8328	0.8190	0.8683	0.9457	0.8026	0.244491
2.00	282	152	0.8509	0.8329	0.8819	0.9338	0.8355	0.229696
4.00	282	152	0.8707	0.8520	0.8819	0.9338	0.8355	0.229696

Label: bus

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849514
1.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849514
2.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849514
4.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849514

Label: bicycle

Threshold	Predict_num	Groundtruth_num	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	2	19	nan	nan	nan	nan
1.00	2	19	nan	nan	nan	nan
2.00	2	19	nan	nan	nan	nan
4.00	2	19	nan	nan	nan	nan

Label: pedestrian

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
0.50	324	123	0.3065	0.2339	0.5123	0.4506	0.5935	0.225948
1.00	324	123	0.3065	0.2339	0.5123	0.4506	0.5935	0.225948
2.00	324	123	0.3201	0.2452	0.5193	0.4568	0.6016	0.225948
4.00	324	123	0.3529	0.2698	0.5488	0.4390	0.7317	0.181541

Summary:

Label	Predict_num	GT_nums	Thresholds	mean AP	APs	Mean APH	APHs
car	586	337	0.50/1.00/2.00/4.00	0.8381	0.7896 / 0.8421 / 0.8604 / 0.8604	0.819	0.7718 / 0.8235 / 0.8412 / 0.8412
truck	282	152	0.50/1.00/2.00/4.00	0.7756	0.5480 / 0.8328 / 0.8509 / 0.8707	0.760	0.5375 / 0.8190 / 0.8329 / 0.8520
bus	23	19	0.50/1.00/2.00/4.00	0.9979	0.9979 / 0.9979 / 0.9979 / 0.9979	0.989	0.9889 / 0.9889 / 0.9889 / 0.9889
bicycle	2	19	0.50/1.00/2.00/4.00	0.0000	0.0000 / 0.0000 / 0.0000 / 0.0000	0.000	0.0000 / 0.0000 / 0.0000 / 0.0000
pedestrian	324	123	0.50/1.00/2.00/4.00	0.3215	0.3065 / 0.3065 / 0.3201 / 0.3529	0.246	0.2339 / 0.2339 / 0.2452 / 0.2698

mAP: 0.6097, mAPH: 0.5846 (Plane Distance)

Label: car

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	586	337	0.8597	0.8407	0.8928	0.9688	0.8279	0.323492
4.00	586	337	0.8604	0.8412	0.8928	0.9688	0.8279	0.323492

Label: truck

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	282	152	0.8509	0.8329	0.8819	0.9338	0.8355	0.229696
4.00	282	152	0.8568	0.8388	0.8819	0.9338	0.8355	0.229696

Label: bus

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849514
4.00	23	19	0.9979	0.9889	1.0000	1.0000	1.0000	0.849514

Label: bicycle

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	2	19	0.0000	0.0000	nan	nan	nan	nan
4.00	2	19	0.0000	0.0000	nan	nan	nan	nan

Label: pedestrian

Threshold	Predict_num	Groundtruth_num	AP	APH	max_f1	optimal_recall	optimal_precision	optimal_conf
2.00	324	123	0.3201	0.2452	0.5193	0.4568	0.6016	0.225948
4.00	324	123	0.3529	0.2698	0.5488	0.4390	0.7317	0.181541

Summary:

Label	Predict_num	GT_nums	Thresholds	mean AP	APs	Mean APH	APHs
car	586	337	2.00/4.00	0.8601	0.8597 / 0.8604	0.841	0.8407 / 0.8412
truck	282	152	2.00/4.00	0.8539	0.8509 / 0.8568	0.836	0.8329 / 0.8388
bus	23	19	2.00/4.00	0.9979	0.9979 / 0.9979	0.989	0.9889 / 0.9889
bicycle	2	19	2.00/4.00	0.0000	0.0000 / 0.0000	0.000	0.0000 / 0.0000
pedestrian	324	123	2.00/4.00	0.3365	0.3201 / 0.3529	0.258	0.2452 / 0.2698

Copilot

Pull request overview

This PR integrates CenterPoint into a unified deployment pipeline by creating a new project bundle under deployment/projects/centerpoint/ and removing legacy deployment scripts. The integration enables PyTorch, ONNX, and TensorRT export/evaluation workflows through a componentized multi-file ONNX approach.

Key Changes:

Centralizes CenterPoint deployment under a unified CLI (deployment.cli.main centerpoint)
Implements multi-file ONNX export (voxel encoder + backbone/head components) with per-component TensorRT engine conversion
Removes legacy deployment entrypoints (projects/CenterPoint/scripts/deploy.py, projects/CenterPoint/runners/deployment_runner.py, projects/CenterPoint/models/detectors/centerpoint_onnx.py)

Reviewed changes

Copilot reviewed 26 out of 26 changed files in this pull request and generated 9 comments.

Show a summary per file

File	Description
`projects/CenterPoint/scripts/deploy.py`	Removed legacy deployment script
`projects/CenterPoint/runners/deployment_runner.py`	Removed legacy runner implementation
`projects/CenterPoint/models/detectors/centerpoint_onnx.py`	Removed legacy ONNX detector variant
`projects/CenterPoint/models/__init__.py`	Removed ONNX model imports/exports
`projects/CenterPoint/README.md`	Updated with unified deployment CLI usage
`projects/CenterPoint/Dockerfile`	Added deployment dependencies (onnxruntime-gpu, tensorrt-cu12)
`deployment/projects/centerpoint/runner.py`	New unified deployment runner
`deployment/projects/centerpoint/pipelines/*.py`	Backend-specific inference pipelines (PyTorch/ONNX/TensorRT)
`deployment/projects/centerpoint/export/*.py`	ONNX/TensorRT export pipelines with component extraction
`deployment/projects/centerpoint/onnx_models/*.py`	Relocated ONNX model variants with import path fixes
`deployment/projects/centerpoint/*.py`	Core deployment components (entrypoint, evaluator, data_loader, model_loader)
`deployment/projects/centerpoint/config/deploy_config.py`	Centralized deployment configuration
`deployment/projects/centerpoint/cli.py`	CLI flag registration (`--rot-y-axis-reference`)

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

deployment/projects/centerpoint/onnx_models/pillar_encoder_onnx.py

deployment/projects/centerpoint/onnx_models/centerpoint_head_onnx.py

deployment/projects/centerpoint/onnx_models/centerpoint_onnx.py

deployment/projects/centerpoint/pipelines/tensorrt.py

deployment/projects/centerpoint/model_loader.py

deployment/projects/centerpoint/export/tensorrt_export_pipeline.py

deployment/projects/centerpoint/evaluator.py

deployment/projects/centerpoint/data_loader.py

projects/CenterPoint/README.md

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

…x.py Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

…nnx.py Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

KSeangTan · 2026-01-29T10:15:06Z

deployment/core/config/base_config.py

-        model_inputs: Tuple[TensorRTModelInputConfig, ...] = tuple(
-            TensorRTModelInputConfig.from_dict(item) for item in model_inputs_raw
-        )
+    def from_dict(cls, config_dict: Mapping[str, Any]) -> "TensorRTConfig":


use from __future__ import annotations for self-reference without ""

KSeangTan · 2026-01-29T10:15:52Z

deployment/core/config/base_config.py

+            max_workspace_size=config_dict.get("max_workspace_size", DEFAULT_WORKSPACE_SIZE),
        )

    def get_precision_policy(self) -> str:


Adding @property is cleaner, and remove get_

Thanks, fixed in another PR in b75687c

deployment/core/config/base_config.py

KSeangTan · 2026-01-29T10:20:57Z

deployment/core/config/base_config.py

        self.export_config = ExportConfig.from_dict(deploy_cfg.get("export", {}))
        self.runtime_config = RuntimeConfig.from_dict(deploy_cfg.get("runtime_io", {}))
-        self.backend_config = BackendConfig.from_dict(deploy_cfg.get("backend_config", {}))
+        self.tensorrt_config = TensorRTConfig.from_dict(deploy_cfg.get("tensorrt_config", {}) or {})


Do TensorRTConfig.from_dict return {}? Is so, we dont need or {}

Thanks, fixed in 152ff54

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

vividf · 2026-02-02T16:23:31Z

Close this PR and seperate them to two new PRs. One for refactor and one for integrate centerpoint.
@KSeangTan
Please review them inorder, thanks

vividf requested a review from Copilot December 25, 2025 08:42

vividf self-assigned this Dec 25, 2025

Copilot AI reviewed Dec 25, 2025

View reviewed changes

vividf requested review from KSeangTan and yamsam January 9, 2026 04:00

vividf marked this pull request as ready for review January 9, 2026 04:00

vividf force-pushed the feat/new_deployment_and_evaluation_pipeline branch from 68d1700 to 7ac5b8b Compare January 29, 2026 08:08

vividf force-pushed the feat/centerpoint_integration branch from 0c61463 to a9cd830 Compare January 29, 2026 08:10

vividf and others added 15 commits January 29, 2026 18:39

feat: integrate centerpoint to new deployment framework

3f9e293

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

Update projects/CenterPoint/README.md

11780c9

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Update deployment/projects/centerpoint/pipelines/tensorrt.py

8e47c76

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Update deployment/projects/centerpoint/onnx_models/pillar_encoder_onn…

bd6e54e

…x.py Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Update deployment/projects/centerpoint/onnx_models/centerpoint_head_o…

8cdedff

…nnx.py Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

chore: removed comment

d4d5553

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

chore: clean up centerpoint export pipeline

52f433d

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

chore: rename backend config to trt config

4a26d9e

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

refactor: refactor deploy config

1ea169c

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

chore: clean code

b52dc4f

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

chore: integrate centerpoint with new component structure

ec9e97f

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

chore: fix kwargs

6dfdc59

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

chore: update metrics output

359bcbb

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

chore: clean code

f7a90c1

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

chore: clean up kwargs

97d977b

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

vividf force-pushed the feat/centerpoint_integration branch from a9cd830 to 97d977b Compare January 29, 2026 09:40

KSeangTan reviewed Jan 29, 2026

View reviewed changes

deployment/core/config/base_config.py Show resolved Hide resolved

KSeangTan reviewed Jan 29, 2026

View reviewed changes

deployment/core/config/base_config.py Show resolved Hide resolved

KSeangTan reviewed Jan 29, 2026

View reviewed changes

chore: temp fix for t4metric2

a9ecfdd

Signed-off-by: vividf <yihsiang.fang@tier4.jp>

vividf closed this Feb 2, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(centerpoint): integrate CenterPoint into unified deployment pipeline#161

feat(centerpoint): integrate CenterPoint into unified deployment pipeline#161
vividf wants to merge 16 commits intotier4:feat/new_deployment_and_evaluation_pipelinefrom
vividf:feat/centerpoint_integration

vividf commented Dec 25, 2025 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

KSeangTan Jan 29, 2026

Uh oh!

KSeangTan Jan 29, 2026 •

edited

Loading

Uh oh!

vividf Feb 3, 2026

Uh oh!

Uh oh!

Uh oh!

KSeangTan Jan 29, 2026

Uh oh!

vividf Feb 3, 2026

Uh oh!

vividf commented Feb 2, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

vividf commented Dec 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Change points

How to run

Exported ONNX (Same)

Evaluation result (Same for Deployment and Test)

Test with test.py

Test with Deployment pipeline

PYTORCH Results:

ONNX Results:

TENSORRT Results:

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

KSeangTan Jan 29, 2026

Choose a reason for hiding this comment

Uh oh!

KSeangTan Jan 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vividf Feb 3, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

KSeangTan Jan 29, 2026

Choose a reason for hiding this comment

Uh oh!

vividf Feb 3, 2026

Choose a reason for hiding this comment

Uh oh!

vividf commented Feb 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

vividf commented Dec 25, 2025 •

edited

Loading

KSeangTan Jan 29, 2026 •

edited

Loading

vividf commented Feb 2, 2026 •

edited

Loading