Convert Tensoflow V2 model ( pb format) to UFF

[parham.khojasteh]

I’m trying to use inference API (detectnet.py) to run an object detection model ( “efficientdet_d1” model which was trained by Tensorflow object detection API 2 on a local machine) on jetson Nano. I gather that UFF conversation (uff.from_tensorflow_frozen_model) doesn’t support TF2 for direct conversion, so I’m following this pathway:

PB format—> ONNX format —> UFF

I could successfully generate ONNX format. However, using “trtexec” led to rasing below issue:

I was wondering if you have a solution.

Unsupported ONNX data type: UINT8 (2)
ERROR: ModelImporter.cpp:54 In function importInput:
[8] Assertion failed: convert_dtype(onnx_tensor_type.elem_type(), &trt_dtype)
[10/16/2020-14:30:38] [E] Failed to parse onnx file
[10/16/2020-14:30:38] [E] Parsing model failed
[10/16/2020-14:30:38] [E] Engine could not be created
&&&& FAILED TensorRT.trtexec # /usr/src/tensorrt/bin/trtexec --onnx=/media/pk/Data/Project/Project/Venice/Maet_stage1/efficientdet_d1_coco17_tpu-32/6class/ONNX/model_TRT.onnx

[AastaLLL]

Hi,

TensorRT can support ONNX format directly, you don’t need to convert it back to UFF.

To fix the type non-supported issue, please check below topic for information:

Exporting Tensorflow models to Jetson Nano

Hi, Elviron The root cause is onnx expects input image to be INT8 but TensorRT use Float32. To solve this issue, you can modify the input data format of ONNX with our graphsurgeon API directly. 1. Install ONNX Graphsurgeon API $ sudo apt-get install python3-pip libprotobuf-dev protobuf-compiler $ git clone https://github.com/NVIDIA/TensorRT.git $ cd TensorRT/tools/onnx-graphsurgeon/ $ make install 2. Modify your model import onnx_graphsurgeon as gs import onnx import numpy as np graph = g…

Thanks.

[parham.khojasteh]

Thanks for your prompt response.

Could you please let me know how I can load a ONNX format with ‘jetson.inference.detectNet’. It’s only able to load a UFF format. When I try to load a ONNX model it gives the error below : ( IS there any tutorial explaining this?

detectNet -- loading detection network model from:
          -- prototxt     NULL
          -- model        /home/bl/Desktop/Workplace/projects/Meat_stag1/efficientdet_d1_coco17_tpu-32/ONNX/model_tf2.onnx
          -- input_blob   'data'
          -- output_cvg   'coverage'
          -- output_bbox  'bboxes'
          -- mean_pixel   0.000000
          -- mean_binary  NULL
          -- class_labels NULL
          -- threshold    0.500000
          -- batch_size   1

[TRT]    TensorRT version 7.1.3
[TRT]    loading NVIDIA plugins...
[TRT]    Registered plugin creator - ::GridAnchor_TRT version 1
[TRT]    Registered plugin creator - ::NMS_TRT version 1
[TRT]    Registered plugin creator - ::Reorg_TRT version 1
[TRT]    Registered plugin creator - ::Region_TRT version 1
[TRT]    Registered plugin creator - ::Clip_TRT version 1
[TRT]    Registered plugin creator - ::LReLU_TRT version 1
[TRT]    Registered plugin creator - ::PriorBox_TRT version 1
[TRT]    Registered plugin creator - ::Normalize_TRT version 1
[TRT]    Registered plugin creator - ::RPROI_TRT version 1
[TRT]    Registered plugin creator - ::BatchedNMS_TRT version 1
[TRT]    Could not register plugin creator -  ::FlattenConcat_TRT version 1
[TRT]    Registered plugin creator - ::CropAndResize version 1
[TRT]    Registered plugin creator - ::DetectionLayer_TRT version 1
[TRT]    Registered plugin creator - ::Proposal version 1
[TRT]    Registered plugin creator - ::ProposalLayer_TRT version 1
[TRT]    Registered plugin creator - ::PyramidROIAlign_TRT version 1
[TRT]    Registered plugin creator - ::ResizeNearest_TRT version 1
[TRT]    Registered plugin creator - ::Split version 1
[TRT]    Registered plugin creator - ::SpecialSlice_TRT version 1
[TRT]    Registered plugin creator - ::InstanceNormalization_TRT version 1
[TRT]    detected model format - ONNX  (extension '.onnx')
[TRT]    desired precision specified for GPU: FASTEST
[TRT]    requested fasted precision for device GPU without providing valid calibrator, disabling INT8
[TRT]    native precisions detected for GPU:  FP32, FP16
[TRT]    selecting fastest native precision for GPU:  FP16
[TRT]    attempting to open engine cache file /home/bl/Desktop/Workplace/projects/Meat_stag1/efficientdet_d1_coco17_tpu-32/ONNX/model_tf2.onnx.1.1.7103.GPU.FP16.engine
[TRT]    cache file not found, profiling network model on device GPU
[TRT]    device GPU, loading /usr/bin/ /home/bl/Desktop/Workplace/projects/Meat_stag1/efficientdet_d1_coco17_tpu-32/ONNX/model_tf2.onnx
----------------------------------------------------------------
Input filename:   /home/bl/Desktop/Workplace/projects/Meat_stag1/efficientdet_d1_coco17_tpu-32/ONNX/model_tf2.onnx
ONNX IR version:  0.0.6
Opset version:    11
Producer name:    tf2onnx
Producer version: 1.8.0
Domain:           
Model version:    0
Doc string:       
----------------------------------------------------------------
[TRT]    Plugin creator already registered - ::GridAnchor_TRT version 1
[TRT]    Plugin creator already registered - ::NMS_TRT version 1
[TRT]    Plugin creator already registered - ::Reorg_TRT version 1
[TRT]    Plugin creator already registered - ::Region_TRT version 1
[TRT]    Plugin creator already registered - ::Clip_TRT version 1
[TRT]    Plugin creator already registered - ::LReLU_TRT version 1
[TRT]    Plugin creator already registered - ::PriorBox_TRT version 1
[TRT]    Plugin creator already registered - ::Normalize_TRT version 1
[TRT]    Plugin creator already registered - ::RPROI_TRT version 1
[TRT]    Plugin creator already registered - ::BatchedNMS_TRT version 1
[TRT]    Could not register plugin creator -  ::FlattenConcat_TRT version 1
[TRT]    Plugin creator already registered - ::CropAndResize version 1
[TRT]    Plugin creator already registered - ::DetectionLayer_TRT version 1
[TRT]    Plugin creator already registered - ::Proposal version 1
[TRT]    Plugin creator already registered - ::ProposalLayer_TRT version 1
[TRT]    Plugin creator already registered - ::PyramidROIAlign_TRT version 1
[TRT]    Plugin creator already registered - ::ResizeNearest_TRT version 1
[TRT]    Plugin creator already registered - ::Split version 1
[TRT]    Plugin creator already registered - ::SpecialSlice_TRT version 1
[TRT]    Plugin creator already registered - ::InstanceNormalization_TRT version 1
Unsupported ONNX data type: UINT8 (2)
ERROR: input_tensor:0:188 In function importInput:
[8] Assertion failed: convertDtype(onnxDtype.elem_type(), &trtDtype)
[TRT]    failed to parse ONNX model '/home/bl/Desktop/Workplace/projects/Meat_stag1/efficientdet_d1_coco17_tpu-32/ONNX/model_tf2.onnx'
[TRT]    device GPU, failed to load /home/bl/Desktop/Workplace/projects/Meat_stag1/efficientdet_d1_coco17_tpu-32/ONNX/model_tf2.onnx
[TRT]    detectNet -- failed to initialize.
jetson.inference -- detectNet failed to load network
Traceback (most recent call last):
  File "detectnet-camera.py", line 49, in <module>
    net = jetson.inference.detectNet(opt.network, sys.argv, opt.threshold)
Exception: jetson.inference -- detectNet failed to load network

[parham.khojasteh]

Update on the last progress:
@AastaLLL I followed the instruction for changing INT8 to Float32, I’ve got this error using modified ONNX format ( withFloat32) on both ‘jetson.inference.detectNet’ and “trtexec” execution:

ERROR: builtin_op_importers.cpp:1554 In function importIf:
[8] Assertion failed: cond.is_weights() && cond.weights().count() == 1 && “If condition must be a initializer!”
[TRT] failed to parse ONNX model ‘/home/bl/Desktop/Workplace/projects/Meat_stag1/efficientdet_d1_coco17_tpu-32/ONNX/modifed_model_TRT.onnx’
[TRT] device GPU, failed to load /home/bl/Desktop/Workplace/projects/Meat_stag1/efficientdet_d1_coco17_tpu-32/ONNX/modifed_model_TRT.onnx
[TRT] detectNet – failed to initialize.
jetson.inference – detectNet failed to load network
Traceback (most recent call last):
File “detectnet-camera.py”, line 49, in
net = jetson.inference.detectNet(opt.network, sys.argv, opt.threshold)
Exception: jetson.inference – detectNet failed to load network

[AastaLLL]

Hi,

Could you run the modified model via trtexec with --verbose and share the log with us?

Thanks.

[parham.khojasteh]

Thanks @AastaLLL.

This is the output log running with --verbose

/usr/src/tensorrt/bin/trtexec --onnx=/media/pk/Data/Project/Project/Venice/Maet_stage1/efficientdet_d1_coco17_tpu-32/6class/ONNX/modifed_model_TRT.onnx --verbose
&&&& RUNNING TensorRT.trtexec # /usr/src/tensorrt/bin/trtexec --onnx=/media/pk/Data/Project/Project/Venice/Maet_stage1/efficientdet_d1_coco17_tpu-32/6class/ONNX/modifed_model_TRT.onnx --verbose
[10/17/2020-14:56:09] [I] === Model Options ===
[10/17/2020-14:56:09] [I] Format: ONNX
[10/17/2020-14:56:09] [I] Model: /media/pk/Data/Project/Project/Venice/Maet_stage1/efficientdet_d1_coco17_tpu-32/6class/ONNX/modifed_model_TRT.onnx
[10/17/2020-14:56:09] [I] Output:
[10/17/2020-14:56:09] [I] === Build Options ===
[10/17/2020-14:56:09] [I] Max batch: 1
[10/17/2020-14:56:09] [I] Workspace: 16 MB
[10/17/2020-14:56:09] [I] minTiming: 1
[10/17/2020-14:56:09] [I] avgTiming: 8
[10/17/2020-14:56:09] [I] Precision: FP32
[10/17/2020-14:56:09] [I] Calibration:
[10/17/2020-14:56:09] [I] Safe mode: Disabled
[10/17/2020-14:56:09] [I] Save engine:
[10/17/2020-14:56:09] [I] Load engine:
[10/17/2020-14:56:09] [I] Inputs format: fp32:CHW
[10/17/2020-14:56:09] [I] Outputs format: fp32:CHW
[10/17/2020-14:56:09] [I] Input build shapes: model
[10/17/2020-14:56:09] [I] === System Options ===
[10/17/2020-14:56:09] [I] Device: 0
[10/17/2020-14:56:09] [I] DLACore:
[10/17/2020-14:56:09] [I] Plugins:
[10/17/2020-14:56:09] [I] === Inference Options ===
[10/17/2020-14:56:09] [I] Batch: 1
[10/17/2020-14:56:09] [I] Iterations: 10 (200 ms warm up)
[10/17/2020-14:56:09] [I] Duration: 10s
[10/17/2020-14:56:09] [I] Sleep time: 0ms
[10/17/2020-14:56:09] [I] Streams: 1
[10/17/2020-14:56:09] [I] Spin-wait: Disabled
[10/17/2020-14:56:09] [I] Multithreading: Enabled
[10/17/2020-14:56:09] [I] CUDA Graph: Disabled
[10/17/2020-14:56:09] [I] Skip inference: Disabled
[10/17/2020-14:56:09] [I] Input inference shapes: model
[10/17/2020-14:56:09] [I] === Reporting Options ===
[10/17/2020-14:56:09] [I] Verbose: Enabled
[10/17/2020-14:56:09] [I] Averages: 10 inferences
[10/17/2020-14:56:09] [I] Percentile: 99
[10/17/2020-14:56:09] [I] Dump output: Disabled
[10/17/2020-14:56:09] [I] Profile: Disabled
[10/17/2020-14:56:09] [I] Export timing to JSON file:
[10/17/2020-14:56:09] [I] Export profile to JSON file:
[10/17/2020-14:56:09] [I]
[10/17/2020-14:56:09] [V] [TRT] Plugin Creator registration succeeded - GridAnchor_TRT
[10/17/2020-14:56:09] [V] [TRT] Plugin Creator registration succeeded - NMS_TRT
[10/17/2020-14:56:09] [V] [TRT] Plugin Creator registration succeeded - Reorg_TRT
[10/17/2020-14:56:09] [V] [TRT] Plugin Creator registration succeeded - Region_TRT
[10/17/2020-14:56:09] [V] [TRT] Plugin Creator registration succeeded - Clip_TRT
[10/17/2020-14:56:09] [V] [TRT] Plugin Creator registration succeeded - LReLU_TRT
[10/17/2020-14:56:09] [V] [TRT] Plugin Creator registration succeeded - PriorBox_TRT
[10/17/2020-14:56:09] [V] [TRT] Plugin Creator registration succeeded - Normalize_TRT
[10/17/2020-14:56:09] [V] [TRT] Plugin Creator registration succeeded - RPROI_TRT
[10/17/2020-14:56:09] [V] [TRT] Plugin Creator registration succeeded - BatchedNMS_TRT
[10/17/2020-14:56:09] [V] [TRT] Plugin Creator registration succeeded - FlattenConcat_TRT

Input filename: /media/pk/Data/Project/Project/Venice/Maet_stage1/efficientdet_d1_coco17_tpu-32/6class/ONNX/modifed_model_TRT.onnx
ONNX IR version: 0.0.7
Opset version: 11
Producer name:
Producer version:
Domain:
Model version: 0
Doc string:

WARNING: ONNX model has a newer ir_version (0.0.7) than this parser was built against (0.0.3).
[10/17/2020-14:56:10] [E] [TRT] Parameter check failed at: …/builder/Network.cpp::addInput::671, condition: isValidDims(dims, hasImplicitBatchDimension())
ERROR: ModelImporter.cpp:80 In function importInput:
[8] Assertion failed: *tensor = importer_ctx->network()->addInput( input.name().c_str(), trt_dtype, trt_dims)
[10/17/2020-14:56:10] [E] Failed to parse onnx file
[10/17/2020-14:56:10] [E] Parsing model failed
[10/17/2020-14:56:10] [E] Engine could not be created
&&&& FAILED TensorRT.trtexec # /usr/src/tensorrt/bin/trtexec --onnx=/media/pk/Data/Project/Project/Venice/Maet_stage1/efficientdet_d1_coco17_tpu-32/6class/ONNX/modifed_model_TRT.onnx --verbose

[AastaLLL]

Hi,

Sorry for the late update.

Could you try the suggestion in the below comment first:

tensorrt inference error while load onnx model

Hi, There may be a couple fixes for the output error. First can you check that the model was parsed correctly? parser->parse… should return True/False depending on whether it parsed the model successfully. Example: TensorRT/sampleINT8API.cpp at 07ed9b57b1ff7c24664388e5564b17f7ce2873e5 · NVIDIA/TensorRT · GitHub If the model was parsed successfully, then the output error has a pretty simple fix. Either mark the output with output_names=… in torch.onnx.export() when exporting your model to ONNX…

Thanks.