deep learning classification results for images in matlab are different from the application compiled by GPU coder on Nvidia Jetson Xavier NX

1. using the attached example (modified from the matlab example) MATLAB\Examples\R2022a\deeplearning_shared\DeployAndClassifyWebcamImagesOnJetsonTX2Example.

2. to call resnet50(), classify the pepper.jpg file :

3. compile the attached codes using GPU coder. Then execute the file to classify pepper.jpg file. It shows different classification results from the Matlab.

4. The deep learning classification results in Matlab id different from the exectuable file running on nvidia JETSON device (which exectuable file is compiled by GPU coder) .

What is wrong?

[Nathan Malimban]

Hi Liwei,

I do not see any attachments in your original post. Could you please attach a script that does the following:

1. Runs the MATLAB code.
2. Generates and compiles the application.
3. Runs the application.
4. Compares the MATLAB classification results and the generated application's classification results.

Thank you!

Nathan

Answer 1

[Liwei]

MATLAB RESULTS for pepper:

top5labels =

5×1 cell array

{'bell pepper' }

{'cucumber' }

{'acorn squash'}

{'lemon' }

{'zucchini' }

Nvidia Xavier NX results (top tw labels):

Bannister 19.2%

Pole 15.8%

The deep learning prediction results in matlab are totally different from Nivida Xavier NX results. This exists in MATLAB 2022a and Matlab2022b (latest release). Would you please help fix the issue as soon as possible?

thank you!

[Nathan Malimban]

Hi Liwei,

Thanks for this info! Could you also give me some information about your Jetson board?

When you run

hwobj= jetson('liwei-nvidia','liwei','888');

what is the output? For example, my Jetson info output looks like this:

Checking for CUDA availability on the Target...
Checking for 'nvcc' in the target system path...
Checking for cuDNN library availability on the Target...
Checking for TensorRT library availability on the Target...
Checking for prerequisite libraries is complete.
Gathering hardware details...
Checking for third-party library availability on the Target...
Gathering hardware details is complete.
 Board name              : NVIDIA Jetson Xavier NX
 CUDA Version            : 10.2
 cuDNN Version           : 8.2
 TensorRT Version        : 8.2
 GStreamer Version       : 1.14.5
 V4L2 Version            : 1.14.2-1
 SDL Version             : 1.2
 OpenCV Version          : 4.1.1
 Available Webcams       :  
 Available GPUs          : Xavier
 Available Digital Pins  : 7  11  12  13  15  16  18  19  21  22  23  24  26  29  31  32  33  35  36  37  38  40

Answer 2

[Liwei]

Here it is.

Gathering hardware details is complete.

Board name : NVIDIA Jetson Xavier NX

CUDA Version : 10.2

cuDNN Version : 8.0

TensorRT Version : 7.1

GStreamer Version : 1.14.5

V4L2 Version : 1.14.2-1

SDL Version :

OpenCV Version : 4.1.1

Available Webcams :

Available GPUs : Xavier

Available Digital Pins : 7 11 12 13 15 16 18 19 21 22 23 24 26 29 31 32 33 35 36 37 38 40