TDA4VH-Q1: Are official PointPillars ONNX artifacts available for SDK 9.2 or 10 on AM69A, and will upgrading to SDK 10 resolve current artifact generation issues?

Muhammad Umair Khan

Part Number: TDA4VH-Q1
Other Parts Discussed in Thread: AM69A

Tool/software:

Dear TI Support Team,

We are working with the AM69A board and currently using SDK 9.2 for model deployment and real-time inference. Our goal is to run the PointPillars model on this platform. After research, we have encountered several uncertainties regarding the availability and compatibility of ONNX artifacts for PointPillars, and we would appreciate your clarification on the following points:

1. Availability of PointPillars Artifacts for SDK 9.2 and 10
We have only found a PointPillars ONNX artifact (3dod-7100_onnxrt_kitti_mmdet3d_lidar_point_pillars_10k_496x432_qat-p2_onnx.tar.gz) provided for release 9.0.

Despite reviewing the documentation and model zoo, we have not located updated artifacts for SDK 9.2 or SDK 101.

Question: Are there official PointPillars ONNX artifacts available for SDK 9.2 or SDK 10? If so, could you please point us to their location? Having these would help us verify our inference pipeline and ensure our deployment process is correct.

2. Plans for Future Artifact Releases
Question: If such artifacts are not currently available, does TI have a plan or timeline to release PointPillars artifacts compatible with SDK 9.2 or SDK 10 in the near future?

3. Issues with Artifact Generation from Custom Training
We have trained the PointPillars model using the configuration at:
github.com/.../tidl_hv_pointpillars_secfpn_6x8_160e_kitti-3d-car.py

However, we are unable to generate the necessary artifacts for deployment on the AM69A board.

Question: Are there any known issues or additional steps required to successfully generate artifacts for PointPillars under SDK 9.2 or SDK 10?

4. SDK Upgrade and Issue Resolution
Question: If we upgrade to SDK 10, will this resolve the artifact generation and deployment issues we are experiencing?

2 months ago

0 Christina Kuruvilla 2 months ago

TI__Expert 4100 points

Hello Umair,

Q1 & Q2: Is this what you were looking for? It is for 10.1, but the 9.2 version can be found by changing the branch. https://github.com/TexasInstruments/edgeai-tensorlab/tree/main/edgeai-modelzoo/models/vision/detection_3d/kitti/mmdet3d

Q3 & Q4: Could you share the steps and the logs for the compiling that you are using? Are you using OSRT or TIDLRT flow? I am not aware of any known issues or additional steps, but I need some more information on how to recreate your setup before I can comment better.

Warm regards,

Christina

0 Muhammad Umair Khan 2 months ago in reply to Christina Kuruvilla

Prodigy 10 points

Hi Christina, thank you for your response.

As you suggested, we are using the following:

ONNX model: github.com/.../lidar_point_pillars_10k_496x432_qat-p2.onnx.link

Prototxt file: github.com/.../lidar_point_pillars_10k_496x432.prototxt.link

Our framework is edgeai-benchmark version r9.2:

github.com/.../edgeai-benchmark

We believe all related information can be found in the log below. We have set the debug level to 3.

We encountered three warnings:

- Warning: Requested Input Data Convert Layer is not added to the network. It is currently not optimal.

- Warning: Requested Output Data Convert Layer is not added to the network. It is currently not optimal.

- WARNING: Conv Layer Conv_8's coefficients cannot be found (or do not match) in the coefficient file. Random coefficients will be generated! Only for evaluation usage! Results are all random!

We understand the first two warnings, as model optimization is not applied in this 3D detection experiment. However, the last warning seems abnormal. I found a related discussion here:

e2e.ti.com/.../tda4vm-mnist-model-import-error-tidl

That said, the model runs normally with CPUExecutionProvider, so it doesn't seem to be an issue with a dummy model. The input format should also be correct, as I used the same example you provided. Additionally, I verified the input shape, and it matches the expected dimensions.

Please find the log below:

TARGET_SOC: AM69A

TARGET_MACHINE: pc

DEBUG MODE: false @ user:5678

=> Recommend to use the alternate script run_benchmarks_parallelbash_pc.sh

for import and inference across models in parallel.

TIDL_TOOLS_PATH=/opt/code/tools/AM69A/tidl_tools

LD_LIBRARY_PATH=/opt/code/tools/AM69A/tidl_tools

PYTHONPATH=:

===================================================================

argv: ['./scripts/benchmark_modelzoo.py', 'settings_import_on_pc.yaml', '--target_device', 'AM69A', '--run_inference', 'False']

settings: {'include_files': None, 'pipeline_type': 'accuracy', 'num_frames': 5, 'calibration_frames': 1, 'calibration_iterations': 1, 'configs_path': './configs', 'models_path': '../edgeai-modelzoo/models', 'modelartifacts_path': './work_dirs/modelartifacts/AM69A', 'modelpackage_path': './work_dirs/modelpackage/AM69A', 'datasets_path': './dependencies/datasets', 'target_device': 'AM69A', 'target_machine': 'pc', 'run_suffix': None, 'parallel_devices': 1, 'parallel_processes': 1, 'tensor_bits': 8, 'runtime_options': {'advanced_options:quantization_scale_type': 4}, 'run_import': True, 'run_inference': False, 'run_missing': True, 'detection_threshold': 0.3, 'detection_top_k': 200, 'detection_nms_threshold': None, 'detection_keep_top_k': None, 'save_output': True, 'num_output_frames': 50, 'model_selection': ['3dod-7100'], 'model_shortlist': None, 'model_exclusion': None, 'task_selection': ['detection_3d'], 'runtime_selection': None, 'session_type_dict': {'onnx': 'onnxrt', 'tflite': 'tflitert', 'mxnet': 'tvmdlr'}, 'dataset_type_dict': None, 'dataset_selection': ['kitti_lidar_det_1class'], 'dataset_loading': True, 'config_range': None, 'enable_logging': True, 'verbose': False, 'capture_log': False, 'experimental_models': True, 'rewrite_results': False, 'with_udp': True, 'flip_test': False, 'model_transformation_dict': None, 'report_perfsim': False, 'tidl_offload': True, 'input_optimization': None, 'run_dir_tree_depth': None, 'target_device_preset': True, 'fast_calibration_factor': 0.5, 'param_template_file': None, 'write_results': True, 'settings_file': 'settings_import_on_pc.yaml', 'basic_keys': ['include_files', 'pipeline_type', 'num_frames', 'calibration_frames', 'calibration_iterations', 'configs_path', 'models_path', 'modelartifacts_path', 'modelpackage_path', 'datasets_path', 'target_device', 'target_machine', 'run_suffix', 'parallel_devices', 'parallel_processes', 'tensor_bits', 'runtime_options', 'run_import', 'run_inference', 'run_missing', 'detection_threshold', 'detection_top_k', 'detection_nms_threshold', 'detection_keep_top_k', 'save_output', 'num_output_frames', 'model_selection', 'model_shortlist', 'model_exclusion', 'task_selection', 'runtime_selection', 'session_type_dict', 'dataset_type_dict', 'dataset_selection', 'dataset_loading', 'config_range', 'enable_logging', 'verbose', 'capture_log', 'experimental_models', 'rewrite_results', 'with_udp', 'flip_test', 'model_transformation_dict', 'report_perfsim', 'tidl_offload', 'input_optimization', 'run_dir_tree_depth', 'target_device_preset', 'fast_calibration_factor', 'param_template_file', 'write_results', 'settings_file'], 'dataset_cache': None}

work_dir: ./work_dirs/modelartifacts/AM69A/8bits

Using model configs from Python module: ./configs

INFO:20250627-014540: loading dataset - category:kitti_lidar_det_1class variant:kitti_lidar_det_1class

configs to run: ['3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx']

number of configs: 1

INFO:20250627-014540: parallel_run - parallel_processes:1 parallel_devices=range(0, 1)

TASKS | | 0% 0/1| [< ]

INFO:20250627-014540: starting process on parallel_device - 0 0%| || 0/1 [00:00<?, ?it/s]

INFO:20250627-014540: starting - 3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx

INFO:20250627-014540: model_path - /opt/code/models/pointPillar_1class_qat_v92/lidar_point_pillars_10k_496x432_qat-p2.onnx

INFO:20250627-014540: model_file - /opt/code/work_dirs/modelartifacts/AM69A/8bits/3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx/model/lidar_point_pillars_10k_496x432_qat-p2.onnx

INFO:20250627-014540: quant_file - /opt/code/work_dirs/modelartifacts/AM69A/8bits/3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx/model/lidar_point_pillars_10k_496x432_qat-p2_qparams.prototxt

INFO:20250627-014540: running - 3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx

INFO:20250627-014540: pipeline_config - {'task_type': 'detection_3d', 'dataset_category': 'kitti_lidar_det_1class', 'calibration_dataset': <edgeai_benchmark.datasets.kitti_lidar_det.KittiLidar3D object at 0x7fc3286edfc0>, 'input_dataset': <edgeai_benchmark.datasets.kitti_lidar_det.KittiLidar3D object at 0x7fc3286fe4d0>, 'postprocess': <edgeai_benchmark.postprocess.PostProcessTransforms object at 0x7fc2ade1ab00>, 'preprocess': <edgeai_benchmark.preprocess.PreProcessTransforms object at 0x7fc2ade1abf0>, 'session': <edgeai_benchmark.sessions.onnxrt_session.ONNXRTSession object at 0x7fc2ade1ab90>, 'metric': {'label_offset_pred': None}, 'model_info': {'metric_reference': {'accuracy_ap_3d_moderate%': 76.5}, 'model_shortlist': 30}}

INFO:20250627-014540: import - 3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx - this may take some time...tidl_tools_path = /opt/code/tools/AM69A/tidl_tools

artifacts_folder = /opt/code/work_dirs/modelartifacts/AM69A/8bits/3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx/artifacts

tidl_tensor_bits = 8

debug_level = 3

num_tidl_subgraphs = 16

tidl_denylist =

tidl_denylist_layer_name =

tidl_denylist_layer_type =

tidl_allowlist_layer_name =

model_type =

tidl_calibration_accuracy_level = 64

tidl_calibration_options:num_frames_calibration = 1

tidl_calibration_options:bias_calibration_iterations = 1

mixed_precision_factor = -1.000000

model_group_id = 0

power_of_2_quantization = 3

ONNX QDQ Enabled = 0

enable_high_resolution_optimization = 0

pre_batchnorm_fold = 1

add_data_convert_ops = 3

output_feature_16bit_names_list =

m_params_16bit_names_list =

m_single_core_layers_names_list =

reserved_compile_constraints_flag = 83886080

ti_internal_reserved_1 =

3dod_ssd is meta arch name

point_pillars

Number of OD backbone nodes = 59

Size of odBackboneNodeIds = 59

Supported TIDL layer type --- Clip -- Clip_0

Supported TIDL layer type --- Conv -- Conv_1

Supported TIDL layer type --- Clip -- Clip_2

Supported TIDL layer type --- ReduceMax -- ReduceMax_3

Supported TIDL layer type --- Squeeze -- Squeeze_4

Supported TIDL layer type --- Conv -- Conv_8

Supported TIDL layer type --- Clip -- Clip_9

Supported TIDL layer type --- Conv -- Conv_10

Supported TIDL layer type --- Clip -- Clip_11

Supported TIDL layer type --- Conv -- Conv_12

Supported TIDL layer type --- Clip -- Clip_13

Supported TIDL layer type --- Conv -- Conv_14

Supported TIDL layer type --- Clip -- Clip_15

Supported TIDL layer type --- Conv -- Conv_16

Supported TIDL layer type --- Clip -- Clip_17

Supported TIDL layer type --- Conv -- Conv_18

Supported TIDL layer type --- Clip -- Clip_19

Supported TIDL layer type --- Conv -- Conv_20

Supported TIDL layer type --- Clip -- Clip_21

Supported TIDL layer type --- Conv -- Conv_22

Supported TIDL layer type --- Clip -- Clip_23

Supported TIDL layer type --- Conv -- Conv_24

Supported TIDL layer type --- Clip -- Clip_25

Supported TIDL layer type --- Conv -- Conv_26

Supported TIDL layer type --- Clip -- Clip_27

Supported TIDL layer type --- Conv -- Conv_28

Supported TIDL layer type --- Clip -- Clip_29

Supported TIDL layer type --- Conv -- Conv_30

Supported TIDL layer type --- Clip -- Clip_31

Supported TIDL layer type --- Conv -- Conv_32

Supported TIDL layer type --- Clip -- Clip_33

Supported TIDL layer type --- Conv -- Conv_34

Supported TIDL layer type --- Clip -- Clip_35

Supported TIDL layer type --- Conv -- Conv_36

Supported TIDL layer type --- Clip -- Clip_37

Supported TIDL layer type --- Conv -- Conv_38

Supported TIDL layer type --- Clip -- Clip_39

Supported TIDL layer type --- Resize -- Resize_49

Supported TIDL layer type --- Conv -- Conv_50

Supported TIDL layer type --- Clip -- Clip_51

Supported TIDL layer type --- Conv -- Conv_52

Supported TIDL layer type --- Clip -- Clip_53

Supported TIDL layer type --- Resize -- Resize_43

Supported TIDL layer type --- Conv -- Conv_44

Supported TIDL layer type --- Clip -- Clip_45

Supported TIDL layer type --- Conv -- Conv_46

Supported TIDL layer type --- Clip -- Clip_47

Supported TIDL layer type --- Conv -- Conv_40

Supported TIDL layer type --- Clip -- Clip_41

Supported TIDL layer type --- Concat -- Concat_54

Supported TIDL layer type --- Clip -- Clip_55

Supported TIDL layer type --- Conv -- Conv_60

Supported TIDL layer type --- Clip -- Clip_61

Supported TIDL layer type --- Conv -- Conv_58

Supported TIDL layer type --- Clip -- Clip_59

Supported TIDL layer type --- Conv -- Conv_56

Supported TIDL layer type --- Clip -- Clip_57

Preliminary subgraphs created = 1

Final number of subgraphs created are : 1, - Offloaded Nodes - 61, Total Nodes - 61

INFORMATION -- [TIDL_ResizeLayer] Any resize ratio which is power of 2 and greater than 4 will be placed by combination of 4x4 resize layer and 2x2 resize layer. For example a 8x8 resize will be replaced by 4x4 resize followed by 2x2 resize.

Running runtimes graphviz - /opt/code/tools/AM69A/tidl_tools/tidl_graphVisualiser_runtimes.out /opt/code/work_dirs/modelartifacts/AM69A/8bits/3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx/artifacts/allowedNode.txt /opt/code/work_dirs/modelartifacts/AM69A/8bits/3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx/artifacts/tempDir/graphvizInfo.txt /opt/code/work_dirs/modelartifacts/AM69A/8bits/3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx/artifacts/tempDir/runtimes_visualization.svg

*** In TIDL_createStateImportFunc ***

Compute on node : TIDLExecutionProvider_TIDL_0_0

0, Clip, 3, 1, x.3, 200

1, Conv, 3, 1, 200, 397

2, Clip, 3, 1, 397, 207

3, ReduceMax, 1, 1, 207, 208

4, Squeeze, 1, 1, 208, 209

5, ScatterElements, 3, 1, data.1, 210

6, Reshape, 2, 1, 210, 212

7, Conv, 3, 1, 212, 400

8, Clip, 3, 1, 400, 219

9, Conv, 3, 1, 219, 403

10, Clip, 3, 1, 403, 226

11, Conv, 3, 1, 226, 406

12, Clip, 3, 1, 406, 233

13, Conv, 3, 1, 233, 409

14, Clip, 3, 1, 409, 240

15, Conv, 3, 1, 240, 448

16, Clip, 3, 1, 448, 331

17, Conv, 3, 1, 240, 412

18, Clip, 3, 1, 412, 247

19, Conv, 3, 1, 247, 415

20, Clip, 3, 1, 415, 254

21, Conv, 3, 1, 254, 418

22, Clip, 3, 1, 418, 261

23, Conv, 3, 1, 261, 421

24, Clip, 3, 1, 421, 268

25, Conv, 3, 1, 268, 424

26, Clip, 3, 1, 424, 275

27, Conv, 3, 1, 275, 427

28, Clip, 3, 1, 427, 282

29, Resize, 3, 1, 282, 336

30, Conv, 3, 1, 336, 451

31, Clip, 3, 1, 451, 343

32, Conv, 3, 1, 343, 454

33, Clip, 3, 1, 454, 350

34, Conv, 3, 1, 282, 430

35, Clip, 3, 1, 430, 289

36, Conv, 3, 1, 289, 433

37, Clip, 3, 1, 433, 296

38, Conv, 3, 1, 296, 436

39, Clip, 3, 1, 436, 303

40, Conv, 3, 1, 303, 439

41, Clip, 3, 1, 439, 310

42, Conv, 3, 1, 310, 442

43, Clip, 3, 1, 442, 317

44, Conv, 3, 1, 317, 445

45, Clip, 3, 1, 445, 324

46, Resize, 3, 1, 324, 355

47, Conv, 3, 1, 355, 457

48, Clip, 3, 1, 457, 362

49, Conv, 3, 1, 362, 460

50, Clip, 3, 1, 460, 369

51, Concat, 3, 1, 331, 370

52, Clip, 3, 1, 370, 375

53, Conv, 3, 1, 375, 376

54, Clip, 3, 1, 376, 381

55, Conv, 3, 1, 375, 382

56, Clip, 3, 1, 382, 387

57, Conv, 3, 1, 375, 388

58, Clip, 3, 1, 388, 393

59, Concat, 3, 1, 381, 394

60, Reshape, 2, 1, 394, 396

Input tensor name - x.3

Input tensor name - data.1

Input tensor name - coors

Output tensor name - 396

Graph Domain TO version : 11In TIDL_onnxRtImportInit subgraph_name=396

Layer 0, subgraph id 396, name=393

Layer 1, subgraph id 396, name=381

Layer 2, subgraph id 396, name=387

Layer 3, subgraph id 396, name=x.3

Layer 4, subgraph id 396, name=data.1

Layer 5, subgraph id 396, name=coors

TIDL Meta PipeLine (Proto) File : /opt/code/work_dirs/modelartifacts/AM69A/8bits/3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx/model/lidar_point_pillars_10k_496x432.prototxt

3dod_ssd

point_pillars

In TIDL_runtimesOptimizeNet: LayerIndex = 65, dataIndex = 62

Warning : Requested Input Data Convert Layer is not Added to the network, It is currently not Optimal

Warning : Requested Output Data Convert Layer is not Added to the network, It is currently not Optimal

WARNING: Conv Layer Conv_8's coeff cannot be found(or not match) in coef file, Random coeff will be generated! Only for evaluation usage! Results are all random!

************** Frame index 1 : Running float import *************

In TIDL_runtimesPostProcessNet

In TIDL_runtimesPostProcessNet 1

In TIDL_runtimesPostProcessNet 2

In TIDL_runtimesPostProcessNet 3

INFORMATION: [TIDL_ResizeLayer] Resize_43 Any resize ratio which is power of 2 and greater than 4 will be placed by combination of 4x4 resize layer and 2x2 resize layer. For example a 8x8 resize will be replaced by 4x4 resize followed by 2x2 resize.

INFORMATION: [TIDL_ResizeLayer] Resize_49 Any resize ratio which is power of 2 and greater than 4 will be placed by combination of 4x4 resize layer and 2x2 resize layer. For example a 8x8 resize will be replaced by 4x4 resize followed by 2x2 resize.

****************************************************

** 2 WARNINGS 0 ERRORS **

****************************************************

In TIDL_runtimesPostProcessNet 4

************ in TIDL_subgraphRtCreate ************

The soft limit is 2048

The hard limit is 2048

MEM: Init ... !!!

MEM: Init ... Done !!!

0.0s: VX_ZONE_INIT:Enabled

0.7s: VX_ZONE_ERROR:Enabled

0.8s: VX_ZONE_WARNING:Enabled

0.2154s: VX_ZONE_INIT:[tivxInit:185] Initialization Done !!!

Memory used for ALG_SCRATCH_DATA_BUFF_MEMREC is greater than requested

************ TIDL_subgraphRtCreate done ************

******* In TIDL_subgraphRtInvoke ********

0 1.00000 -55.00000 63.00000 6

3 1.00000 -55.00000 63.00000 6

1 1.00000 0.00000 0.00000 6

4 1.00000 0.00000 0.00000 6

5 1.00000 -55.00000 63.00000 6

6 1.00000 -2.00000 2.00000 6

7 1.00000 -0.19525 2.00000 6

8 1.00000 -0.19525 2.00000 6

9 1.00000 -0.19525 2.00000 6

10 1.00000 -0.19525 2.00000 6

2 1.00000 0.00000 255.00000 5

11 1.00000 -0.19525 2.00000 6

12 1.00000 0.00000 1.00000 6

13 1.00000 0.00000 1.00000 6

INFO:20250627-014640: parallel_run - num_total_tasks:1 len(queued_tasks):0 len(process_dict):1 len(result_list):0

14 1.00000 0.00000 1.00000 6

INFO:20250627-014740: parallel_run - num_total_tasks:1 len(queued_tasks):0 len(process_dict):1 len(result_list):0

15 1.00000 0.00000 1.00000 6

Memory for TIDL_refConv2d accumulator is not sufficient exiting...

TASKS | 100%|██████████|| 1/1 [02:12<00:00, 132.57s/it]

TASKS | 100%|██████████|| 1/1 [02:12<00:00, 132.47s/it]

-------------------------------------------------------------------

argv: ['./scripts/benchmark_modelzoo.py', 'settings_import_on_pc.yaml', '--target_device', 'AM69A', '--run_import', 'False']

settings: {'include_files': None, 'pipeline_type': 'accuracy', 'num_frames': 5, 'calibration_frames': 1, 'calibration_iterations': 1, 'configs_path': './configs', 'models_path': '../edgeai-modelzoo/models', 'modelartifacts_path': './work_dirs/modelartifacts/AM69A', 'modelpackage_path': './work_dirs/modelpackage/AM69A', 'datasets_path': './dependencies/datasets', 'target_device': 'AM69A', 'target_machine': 'pc', 'run_suffix': None, 'parallel_devices': 1, 'parallel_processes': 1, 'tensor_bits': 8, 'runtime_options': {'advanced_options:quantization_scale_type': 4}, 'run_import': False, 'run_inference': True, 'run_missing': True, 'detection_threshold': 0.3, 'detection_top_k': 200, 'detection_nms_threshold': None, 'detection_keep_top_k': None, 'save_output': True, 'num_output_frames': 50, 'model_selection': ['3dod-7100'], 'model_shortlist': None, 'model_exclusion': None, 'task_selection': ['detection_3d'], 'runtime_selection': None, 'session_type_dict': {'onnx': 'onnxrt', 'tflite': 'tflitert', 'mxnet': 'tvmdlr'}, 'dataset_type_dict': None, 'dataset_selection': ['kitti_lidar_det_1class'], 'dataset_loading': True, 'config_range': None, 'enable_logging': True, 'verbose': False, 'capture_log': False, 'experimental_models': True, 'rewrite_results': False, 'with_udp': True, 'flip_test': False, 'model_transformation_dict': None, 'report_perfsim': False, 'tidl_offload': True, 'input_optimization': None, 'run_dir_tree_depth': None, 'target_device_preset': True, 'fast_calibration_factor': 0.5, 'param_template_file': None, 'write_results': True, 'settings_file': 'settings_import_on_pc.yaml', 'basic_keys': ['include_files', 'pipeline_type', 'num_frames', 'calibration_frames', 'calibration_iterations', 'configs_path', 'models_path', 'modelartifacts_path', 'modelpackage_path', 'datasets_path', 'target_device', 'target_machine', 'run_suffix', 'parallel_devices', 'parallel_processes', 'tensor_bits', 'runtime_options', 'run_import', 'run_inference', 'run_missing', 'detection_threshold', 'detection_top_k', 'detection_nms_threshold', 'detection_keep_top_k', 'save_output', 'num_output_frames', 'model_selection', 'model_shortlist', 'model_exclusion', 'task_selection', 'runtime_selection', 'session_type_dict', 'dataset_type_dict', 'dataset_selection', 'dataset_loading', 'config_range', 'enable_logging', 'verbose', 'capture_log', 'experimental_models', 'rewrite_results', 'with_udp', 'flip_test', 'model_transformation_dict', 'report_perfsim', 'tidl_offload', 'input_optimization', 'run_dir_tree_depth', 'target_device_preset', 'fast_calibration_factor', 'param_template_file', 'write_results', 'settings_file'], 'dataset_cache': None}

work_dir: ./work_dirs/modelartifacts/AM69A/8bits

Using model configs from Python module: ./configs

INFO:20250627-014753: loading dataset - category:kitti_lidar_det_1class variant:kitti_lidar_det_1class

configs to run: ['3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx']

number of configs: 1

INFO:20250627-014753: parallel_run - parallel_processes:1 parallel_devices=range(0, 1)

TASKS | | 0% 0/1| [< ]

INFO:20250627-014753: starting process on parallel_device - 0 0%| || 0/1 [00:00<?, ?it/s]

INFO:20250627-014753: starting - 3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx

INFO:20250627-014753: model_path - /opt/code/models/pointPillar_1class_qat_v92/lidar_point_pillars_10k_496x432_qat-p2.onnx

INFO:20250627-014753: model_file - /opt/code/work_dirs/modelartifacts/AM69A/8bits/3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx/model/lidar_point_pillars_10k_496x432_qat-p2.onnx

INFO:20250627-014753: quant_file - /opt/code/work_dirs/modelartifacts/AM69A/8bits/3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx/model/lidar_point_pillars_10k_496x432_qat-p2_qparams.prototxt

INFO:20250627-014753: running - 3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx

INFO:20250627-014753: pipeline_config - {'task_type': 'detection_3d', 'dataset_category': 'kitti_lidar_det_1class', 'calibration_dataset': <edgeai_benchmark.datasets.kitti_lidar_det.KittiLidar3D object at 0x7f7529d11fc0>, 'input_dataset': <edgeai_benchmark.datasets.kitti_lidar_det.KittiLidar3D object at 0x7f7529d224d0>, 'postprocess': <edgeai_benchmark.postprocess.PostProcessTransforms object at 0x7f74af453820>, 'preprocess': <edgeai_benchmark.preprocess.PreProcessTransforms object at 0x7f74af453730>, 'session': <edgeai_benchmark.sessions.onnxrt_session.ONNXRTSession object at 0x7f74af453790>, 'metric': {'label_offset_pred': None}, 'model_info': {'metric_reference': {'accuracy_ap_3d_moderate%': 76.5}, 'model_shortlist': 30}}

INFO:20250627-014753: infer - 3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx - this may take some time...libtidl_onnxrt_EP loaded 0x743c4f0

artifacts_folder = /opt/code/work_dirs/modelartifacts/AM69A/8bits/3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx/artifacts

debug_level = 3

target_priority = 0

max_pre_empt_delay = 340282346638528859811704183484516925440.000000

Final number of subgraphs created are : 1, - Offloaded Nodes - 61, Total Nodes - 61

In TIDL_createStateInfer

Compute on node : TIDLExecutionProvider_TIDL_0_0

************ in TIDL_subgraphRtCreate ************

Invoke : ERROR: Unable to open network file /opt/code/work_dirs/modelartifacts/AM69A/8bits/3dod-7100_onnxrt_models_pointPillar_1class_qat_v92_lidar_point_pillars_10k_496x432_qat-p2_onnx/artifacts/396_tidl_net.bin

TASKS | 100%|██████████|| 1/1 [00:00<00:00, 1.98it/s]

-------------------------------------------------------------------

settings: {'include_files': None, 'pipeline_type': 'accuracy', 'num_frames': 5, 'calibration_frames': 1, 'calibration_iterations': 1, 'configs_path': './configs', 'models_path': '../edgeai-modelzoo/models', 'modelartifacts_path': './work_dirs/modelartifacts/', 'modelpackage_path': './work_dirs/modelpackage/', 'datasets_path': './dependencies/datasets', 'target_device': None, 'target_machine': 'pc', 'run_suffix': None, 'parallel_devices': 1, 'parallel_processes': 1, 'tensor_bits': 8, 'runtime_options': None, 'run_import': True, 'run_inference': True, 'run_missing': True, 'detection_threshold': 0.3, 'detection_top_k': 200, 'detection_nms_threshold': None, 'detection_keep_top_k': None, 'save_output': True, 'num_output_frames': 50, 'model_selection': ['3dod-7100'], 'model_shortlist': None, 'model_exclusion': None, 'task_selection': ['detection_3d'], 'runtime_selection': None, 'session_type_dict': {'onnx': 'onnxrt', 'tflite': 'tflitert', 'mxnet': 'tvmdlr'}, 'dataset_type_dict': None, 'dataset_selection': ['kitti_lidar_det_1class'], 'dataset_loading': True, 'config_range': None, 'enable_logging': True, 'verbose': False, 'capture_log': False, 'experimental_models': True, 'rewrite_results': False, 'with_udp': True, 'flip_test': False, 'model_transformation_dict': None, 'report_perfsim': False, 'tidl_offload': True, 'input_optimization': None, 'run_dir_tree_depth': None, 'target_device_preset': True, 'fast_calibration_factor': None, 'param_template_file': None, 'write_results': True, 'skip_pattern': '_package', 'settings_file': 'settings_import_on_pc.yaml', 'basic_keys': ['include_files', 'pipeline_type', 'num_frames', 'calibration_frames', 'calibration_iterations', 'configs_path', 'models_path', 'modelartifacts_path', 'modelpackage_path', 'datasets_path', 'target_device', 'target_machine', 'run_suffix', 'parallel_devices', 'parallel_processes', 'tensor_bits', 'runtime_options', 'run_import', 'run_inference', 'run_missing', 'detection_threshold', 'detection_top_k', 'detection_nms_threshold', 'detection_keep_top_k', 'save_output', 'num_output_frames', 'model_selection', 'model_shortlist', 'model_exclusion', 'task_selection', 'runtime_selection', 'session_type_dict', 'dataset_type_dict', 'dataset_selection', 'dataset_loading', 'config_range', 'enable_logging', 'verbose', 'capture_log', 'experimental_models', 'rewrite_results', 'with_udp', 'flip_test', 'model_transformation_dict', 'report_perfsim', 'tidl_offload', 'input_optimization', 'run_dir_tree_depth', 'target_device_preset', 'fast_calibration_factor', 'param_template_file', 'write_results', 'skip_pattern', 'settings_file'], 'dataset_cache': None}

results found for 4 models

Report generated at ./work_dirs/modelartifacts/

0 Christina Kuruvilla 2 months ago in reply to Muhammad Umair Khan

TI__Expert 4100 points

Hello,

Which version of TIDL tools are you running? There is a backport compatible one for 9.2 (10_00_07_00) : https://github.com/TexasInstruments/edgeai-tidl-tools/blob/master/docs/version_compatibility_table.md

Please see if the same behavior occurs in this one.

Thanks,

Christina

Processors

Processors forum

TDA4VH-Q1: Are official PointPillars ONNX artifacts available for SDK 9.2 or 10 on AM69A, and will upgrading to SDK 10 resolve current artifact generation issues?