Jetson AGX Orin - Inference on RESNET50 model(ONNX)

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
1

We were running inference on RESNET50 model(resnet50_v1.onnx) but getting error

Error : Invalid argument : Got invalid dimensions for input: input_1 for the following indices
index:0 Got:1 Expected:32. Please fix either the inputs or model

Seems batch_size is hard coded to 1, Is there a way to change BATCH_SIZE from 1 to 32

3

Hi,

Have you run the ONNX model with other frameworks?

Thanks.

4

Hi,

I apologize for not being clear in earlier post.
We have used the base code from the following github branch and run the MLPerf benchmark inference on RESNET50 model(resnet50_v1.onnx), then we are getting the invalid dimension error
github branch - mlcommons/inference/tree/master/vision/classification_and_detection

Error : Invalid argument : Got invalid dimensions for input: input_1 for the following indices
index:0 Got:1 Expected:32. Please fix either the inputs or model

And also we have run the MLPerf benchmark inference with tensorflow framework and different models like resnet50, mobilenet, ssd-mobilenet but our results were not matching with the one published on MLCommons, attaching the results here
ML_perf_results.xlsx (9.6 KB)

Thanks

5

Hi,

You can find more information about MLPerf of Orin below:

Do you use a similar environment?
Thanks.

6

We are using the same environment.

Thanks

7

Hi,

Please noted that the v2.1 results is tested with the CUDA-X AI DP software rather than JetPack.
You can find the software below:

For the benchmark results, have you maximized the Orin performance first?
This can be done via the following command:

$ sudo nvpmodel -m 0
$ sudo jetson_clocks

Thanks.

9

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.