[TensorRT] ERROR: (Unnamed Layer* 0) [Convolution]: at least 5 dimensions are required for input - on Jetson Xavier

Robotics & Edge Computing Jetson & Embedded Systems Jetson Xavier NX
1

I have created an onnx model that successfully with TensorRT 7 on my host machine, but it gives the following error on AGX Xavier.

[TensorRT] ERROR: (Unnamed Layer* 0) [Convolution]: at least 5 dimensions are required for input
Traceback (most recent call last):
File “run.py”, line 49, in
engine = backend.prepare(model, device=‘CUDA:0’)
File “/home/yaak/onnx-tensorrt/onnx_tensorrt/backend.py”, line 200, in prepare
return TensorRTBackendRep(model, device, **kwargs)
File “/home/yaak/onnx-tensorrt/onnx_tensorrt/backend.py”, line 86, in init
raise RuntimeError(msg)
RuntimeError: While parsing node number 1:
builtin_op_importers.cpp:695 In function importBatchNormalization:
[6] Assertion failed: scale_weights.shape == weights_shape

On AGX Xavier, I have followed the instructions at GitHub - onnx/onnx-tensorrt at 6.0-full-dims
and I have also installed the TensorRT 6.0 open source libaries and onnx 1.5.0.

The model is open source (r2plus1d_34_clip32_ft_kinetics_from_ig65m-ade133f1.pth] and is taken from GitHub - moabitcoin/ig65m-pytorch: PyTorch 3D video classification models pre-trained on 65 million Instagram videos

I have converted the .pth to .onnx using torch 1.2.0 (also using the 1.3 gives the same error)
And used the following script to build a trt engine for it:

import onnx
import onnx_tensorrt.backend as backend
import numpy as np

model = onnx.load(“r2plus1d_34_clip32_ft_kinetics_from_ig65m-10f4c3bf.onnx”)
engine = backend.prepare(model, device=‘CUDA:0’)
input_data = np.random.random(size=(1, 3, 32, 224, 224)).astype(np.float32)
output_data = engine.run(input_data)[0]
print(output_data)
print(output_data.shape)

I appreciate it if you can help me resolving this error.

Thanks
Nasim

3

Hi,

It looks like you meet the similar error of this issue:

Would you mind to check if the suggestion also works for you first?
Thanks.

4

Hi AastaLLL,

Yes, I did go through the solutions proposed in this link.
I have installed TensorRT OSS v6.0 as well and have also tried to run the code with explicit batch flag. This is the script that I used with explicit batch flag:

import tensorrt as trt

def build_engine(onnx_path, engine_path):
  TRT_LOGGER = trt.Logger()
  with trt.Builder(TRT_LOGGER) as builder:
    builder.max_workspace_size = 1 << 30 # 1GB
    builder.max_batch_size = 1
    builder.fp16_mode = True

    #with builder.create_network() as network:
    explicit_batch_flag = 1 << int(trt.NetworkDefinitionCreationFlag.EXPLICIT_BATCH)
    with builder.create_network(explicit_batch_flag) as network:
      with trt.OnnxParser(network, TRT_LOGGER) as parser:
        with open(onnx_path, 'rb') as model:
          print('here')
          parser.parse(model.read())
          print('here')
          network.mark_output(network.get_layer(network.num_layers-1).get_output(0))
          engine = builder.build_cuda_engine(network)

      with open(engine_path, "wb") as f:
        f.write(engine.serialize())
  return engine

if __name__ == '__main__':
  build_engine('model.onnx', 'model.engine')

The solution has not worked for me unfortunately. I still get the same error:

[TensorRT] ERROR: (Unnamed Layer* 0) [Convolution]: at least 5 dimensions are required for input
[TensorRT] WARNING: Setting layouts of network and plugin input/output tensors to linear, as 3D operators are found and 3D non-linear IO formats are not supported, yet.
[TensorRT] ERROR: (Unnamed Layer* 0) [Convolution]: at least 5 dimensions are required for input
Traceback (most recent call last):
File “run3.py”, line 26, in
build_engine(‘model.onnx’, ‘model.engine’)
File “run3.py”, line 22, in build_engine
f.write(engine.serialize())
AttributeError: ‘NoneType’ object has no attribute ‘serialize’

5

Hi,

To give a further suggestion, would you mind to share your onnx model with us?
Thanks.

6

The model that I am using is [r2plus1d_34_clip32_ft_kinetics_from_ig65m-10f4c3bf.pb] taken from

Link to the model:
https://github.com/moabitcoin/ig65m-pytorch/releases/download/v1.0.0/r2plus1d_34_clip32_ft_kinetics_from_ig65m-10f4c3bf.pb

The extension should be changed to .onnx instead of .pb (it is a typo). The onnx file is generated using the script at

7

Hi AastaLLL,

Any update on this? Please let me know if you need further information to reproduce the issue.

8

There is no update from you for a period, assuming this is not an issue any more.
Hence we are closing this topic. If need further support, please open a new one. Thanks

Hi,

Sorry for the late update.
Would you mind to share the onnx file with us directly?

Thanks.