(Unnamed Layer* 60) [ElementWise]: elementwise inputs must have same dimensions or follow broadcast rules (input dimensions were [2] and [3])

darshancganji12 · June 16, 2021, 6:29am

Hi Nvidia Team,

I have converted the Pytorch Model(trained on custom data) to ONNX.
And While I am converting the ONNX to TRT I am getting the issue as attached below:

[06/16/2021-10:25:53] [V] [TRT] Parsing node: Concat_32 [Concat]
[06/16/2021-10:25:53] [V] [TRT] Searching for input: 433
[06/16/2021-10:25:53] [V] [TRT] Searching for input: 14507
[06/16/2021-10:25:53] [V] [TRT] Searching for input: 14508
[06/16/2021-10:25:53] [V] [TRT] Concat_32 [Concat] inputs: [433 -> (1)], [14507 -> (1)], [14508 -> (1)], 
[06/16/2021-10:25:53] [V] [TRT] Registering constant layer: 14507 for ONNX initializer: 14507
[06/16/2021-10:25:53] [V] [TRT] Registering constant layer: 14508 for ONNX initializer: 14508
[06/16/2021-10:25:53] [V] [TRT] Registering layer: Concat_32 for ONNX node: Concat_32
[06/16/2021-10:25:53] [V] [TRT] Registering tensor: 436 for ONNX tensor: 436
[06/16/2021-10:25:53] [V] [TRT] Concat_32 [Concat] outputs: [436 -> (3)], 
[06/16/2021-10:25:53] [V] [TRT] Parsing node: Tile_33 [Tile]
[06/16/2021-10:25:53] [V] [TRT] Searching for input: Transformation.GridGenerator.inv_delta_C
[06/16/2021-10:25:53] [V] [TRT] Searching for input: 436
[06/16/2021-10:25:53] [V] [TRT] Tile_33 [Tile] inputs: [Transformation.GridGenerator.inv_delta_C -> (23, 23)], [436 -> (3)], 
[06/16/2021-10:25:53] [V] [TRT] Registering constant layer: Transformation.GridGenerator.inv_delta_C for ONNX initializer: Transformation.GridGenerator.inv_delta_C
[06/16/2021-10:25:53] [E] [TRT] (Unnamed Layer* 60) [ElementWise]: elementwise inputs must have same dimensions or follow broadcast rules (input dimensions were [2] and [3]).
[06/16/2021-10:25:53] [V] [TRT] Registering layer: Tile_33 for ONNX node: Tile_33
[06/16/2021-10:25:53] [E] [TRT] (Unnamed Layer* 60) [ElementWise]: elementwise inputs must have same dimensions or follow broadcast rules (input dimensions were [2] and [3]).
[06/16/2021-10:25:53] [V] [TRT] Registering tensor: 437 for ONNX tensor: 437
[06/16/2021-10:25:53] [V] [TRT] Tile_33 [Tile] outputs: [437 -> ()], 
[06/16/2021-10:25:53] [V] [TRT] Parsing node: Unsqueeze_34 [Unsqueeze]
[06/16/2021-10:25:53] [V] [TRT] Searching for input: 430
[06/16/2021-10:25:53] [V] [TRT] Unsqueeze_34 [Unsqueeze] inputs: [430 -> ()], 
[06/16/2021-10:25:53] [V] [TRT] Original shape: (), unsqueezing to: (1,)
[06/16/2021-10:25:53] [V] [TRT] Registering layer: Unsqueeze_34 for ONNX node: Unsqueeze_34
[06/16/2021-10:25:53] [E] [TRT] (Unnamed Layer* 60) [ElementWise]: elementwise inputs must have same dimensions or follow broadcast rules (input dimensions were [2] and [3]).
[06/16/2021-10:25:53] [V] [TRT] Registering tensor: 440 for ONNX tensor: 440
[06/16/2021-10:25:53] [V] [TRT] Unsqueeze_34 [Unsqueeze] outputs: [440 -> ()], 
[06/16/2021-10:25:53] [V] [TRT] Parsing node: Concat_35 [Concat]
[06/16/2021-10:25:53] [V] [TRT] Searching for input: 440
[06/16/2021-10:25:53] [E] [TRT] (Unnamed Layer* 60) [ElementWise]: elementwise inputs must have same dimensions or follow broadcast rules (input dimensions were [2] and [3]).
[06/16/2021-10:25:53] [V] [TRT] Searching for input: 14509
[06/16/2021-10:25:53] [V] [TRT] Searching for input: 14510
[06/16/2021-10:25:53] [V] [TRT] Concat_35 [Concat] inputs: [440 -> ()], [14509 -> (1)], [14510 -> (1)], 
[06/16/2021-10:25:53] [V] [TRT] Registering constant layer: 14509 for ONNX initializer: 14509
[06/16/2021-10:25:53] [V] [TRT] Registering constant layer: 14510 for ONNX initializer: 14510
[06/16/2021-10:25:53] [E] [TRT] (Unnamed Layer* 60) [ElementWise]: elementwise inputs must have same dimensions or follow broadcast rules (input dimensions were [2] and [3]).
[06/16/2021-10:25:53] [E] [TRT] /home/darshan/onnx-tensorrt/ModelImporter.cpp:703: While parsing node number 35 [Concat -> "443"]:
[06/16/2021-10:25:53] [E] [TRT] /home/darshan/onnx-tensorrt/ModelImporter.cpp:704: --- Begin node ---
[06/16/2021-10:25:53] [E] [TRT] /home/darshan/onnx-tensorrt/ModelImporter.cpp:705: input: "440"
input: "14509"
input: "14510"
output: "443"
name: "Concat_35"
op_type: "Concat"
attribute {
  name: "axis"
  i: 0
  type: INT
}

[06/16/2021-10:25:53] [E] [TRT] /home/darshan/onnx-tensorrt/ModelImporter.cpp:706: --- End node ---
[06/16/2021-10:25:53] [E] [TRT] /home/darshan/onnx-tensorrt/ModelImporter.cpp:709: ERROR: /home/darshan/onnx-tensorrt/onnx2trt_utils.cpp:196 In function convertAxis:
[8] Assertion failed: axis >= 0 && axis < nbDims
[06/16/2021-10:25:53] [E] Failed to parse onnx file
[06/16/2021-10:25:53] [E] Parsing model failed
[06/16/2021-10:25:53] [E] Engine creation failed
[06/16/2021-10:25:53] [E] Engine set up failed

One issue is:(Unnamed Layer* 60) [ElementWise]: elementwise inputs must have same dimensions or follow broadcast rules (input dimensions were [2] and [3]).

Other is: [8] Assertion failed: axis >= 0 && axis < nbDims

Please help me to resolve this issue.

Looking forward to your reply

Thanks,
Darshan

NVES · June 16, 2021, 6:37am

Hi,
Request you to share the ONNX model and the script if not shared already so that we can assist you better.
Alongside you can try few things:

validating your model with the below snippet

check_model.py

import sys
import onnx
filename = yourONNXmodel
model = onnx.load(filename)
onnx.checker.check_model(model).
2) Try running your model with trtexec command.
https://github.com/NVIDIA/TensorRT/tree/master/samples/opensource/trtexec
In case you are still facing issue, request you to share the trtexec “”–verbose"" log for further debugging
Thanks!

darshancganji12 · June 16, 2021, 7:09am

Can I get the better answer? I will get the same message when I file some conversion Issue.

spolisetty · June 17, 2021, 12:22pm

Hi @darshancganji12,

Please let us know which TensorRT version are you using. We recommend you to try on latest TensorRT 8.0 version. And if you still face this issue, we recommend you to please share complete verbose logs and if possible issue reproducible ONNX model(you may DM us also) for better debugging.

Thank you.

darshancganji12 · June 17, 2021, 12:53pm

Hi @spolisetty,

Thank you so much for your kind reply as always.
As requested I am attaching the Log and the ONNX model below:

I used torch 1.8.1 with Opset 12 for converting to ONNX.

Error_log.txt (218.1 KB)

Here is the ONNX model: Text_recognition_1.8_12.onnx - Google Drive

Thanks,
Darshan C G

spolisetty · June 18, 2021, 12:43pm

@darshancganji12,

Looks like you missed to share these details. Could you please let us know have you tried on TensorRT 8.0 version and which TensorRT version are you currently using.

Thank you.

darshancganji12 · June 18, 2021, 12:55pm

Hi @spolisetty,

I am using TRT version: 7.1.3.4. Can I get the Docker image for TRT 8.0?

Thanks,
Darshan

spolisetty · June 18, 2021, 12:58pm

Docker image for TRT 8.0 not yet available. you can download latest NGC container and setup TRT 8.0 by following below doc.
https://ngc.nvidia.com/containers/nvidia:tensorrt

Thank you.

Topic		Replies	Views
Issues with torch.nn.ReflectionPad2d(padding) conversion to TRT engine TensorRT tensorrt , pytorch , onnx	21	4158	February 8, 2022
Trtexec can not convert resnet152 onnx to TRT engine, without prompting error! TensorRT	12	1522	July 22, 2021
TensorRT's OnnxParser problem TensorRT tensorrt	6	2316	October 12, 2021
Running a pytorch network converted to ONNX with TensorRT on the TX2 Jetson TX2	24	8839	October 18, 2021
Problem converting TensorFlow 2-> ONNX model to TensorRT Engine (efficientdet_d0) TensorRT	8	1387	November 17, 2022
tensorrt's onnx parser can't parse the output layer correctly TensorRT	12	4838	November 24, 2021
Onnx with dynamic batch cannot be parsed TensorRT tensorrt	12	1515	August 9, 2021
Cannot convert SSD ONNX model to TensorRT TensorRT tensorrt	15	2345	November 23, 2022
Input length mismatch (onnx conversion to .trt) TensorRT tensorrt , onnx	4	1243	July 13, 2022
Gather node output wrong when converting with TensorRT TensorRT	23	1725	January 10, 2023

(Unnamed Layer* 60) [ElementWise]: elementwise inputs must have same dimensions or follow broadcast rules (input dimensions were [2] and [3])

check_model.py

Related topics