cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Highlighted
Contributor
Contributor
573 Views
Registered: ‎08-02-2019

Deploying model on ZCU102 board

Jump to solution

Hi,

 I have run my tensorflow model on the ZCU102 board using python API's for DPU. After evaluating the model on the board what are the steps for deploying the model onto the FPGA? Also, I am using DNNDK v-3.1.

0 Kudos
1 Solution

Accepted Solutions
Highlighted
Moderator
Moderator
483 Views
Registered: ‎03-27-2013

Hi @jbond1007 ,

 

For custom design I would suggest the following steps:

1. Run demon for DNNDK examples

I recomend these examples because they are using low level API and have more detailed code and can be easiler modified to suit your own design

2. Use DNNDK to training your own model into ELF

Following the DNNDK tutorial. Acutally for now the models we provided are just reference design. We don't provide details of the trainng flow. So for a custom design I suppose you would like to train your own models

3. Following the steps in https://github.com/Xilinx/Edge-AI-Platform-Tutorials/tree/master/docs/DPU-Integration to create your own platform.

Xilinx dev board or your own board.

Best Regards,
Jason
-----------------------------------------------------------------------------------------------
Please mark the Answer as "Accept as solution" if the information provided is helpful.

Give Kudos to a post which you think is helpful and reply oriented.
-----------------------------------------------------------------------------------------------

View solution in original post

0 Kudos
4 Replies
Highlighted
Moderator
Moderator
525 Views
Registered: ‎03-27-2013

Hi @jbond1007 ,

 

I am afraid that I don't quite your question.

If you run your model on ZCU102 I believe that you have deploy the model on FPGA.

If you mean creating a DPU design and compiling PetaLinux images I would suggest you to refer to:

https://github.com/Xilinx/Edge-AI-Platform-Tutorials/tree/master/docs/DPU-Integration

If you mean change your own model(Caffee or TensorFlow) into kernel ELF I would suggest you to refer to the flow in:

https://www.xilinx.com/support/documentation/user_guides/ug1327-dnndk-user-guide.pdf#page=37

Hope this is what you want. :-)

Best Regards,
Jason
-----------------------------------------------------------------------------------------------
Please mark the Answer as "Accept as solution" if the information provided is helpful.

Give Kudos to a post which you think is helpful and reply oriented.
-----------------------------------------------------------------------------------------------
0 Kudos
Highlighted
Contributor
Contributor
520 Views
Registered: ‎08-02-2019

Hi,

Thanks for your reply.

I have successfully generated .elf files for my tensorflow model and also have run it on the board. My question is what should be done to deploy this model in live production environment.

Now I have a python code for interacting with the DPU which was meant for evaluating the model on the board similar to "miniresnet.py" from the samples in the DNNDK files. I have tested the model with it and it ran successfully, but what are the next steps for putting this model in production where live images will be captured and the model classifies them?

0 Kudos
Highlighted
Moderator
Moderator
484 Views
Registered: ‎03-27-2013

Hi @jbond1007 ,

 

For custom design I would suggest the following steps:

1. Run demon for DNNDK examples

I recomend these examples because they are using low level API and have more detailed code and can be easiler modified to suit your own design

2. Use DNNDK to training your own model into ELF

Following the DNNDK tutorial. Acutally for now the models we provided are just reference design. We don't provide details of the trainng flow. So for a custom design I suppose you would like to train your own models

3. Following the steps in https://github.com/Xilinx/Edge-AI-Platform-Tutorials/tree/master/docs/DPU-Integration to create your own platform.

Xilinx dev board or your own board.

Best Regards,
Jason
-----------------------------------------------------------------------------------------------
Please mark the Answer as "Accept as solution" if the information provided is helpful.

Give Kudos to a post which you think is helpful and reply oriented.
-----------------------------------------------------------------------------------------------

View solution in original post

0 Kudos
Highlighted
Moderator
Moderator
410 Views
Registered: ‎03-27-2013

Hi @jbond1007 ,

 

For "live images will be captured and the model classifies them" I believe that there should be custom design input here.

And we do have some references but most of them are written in C code like this web camera demo:

https://github.com/Xilinx/Edge-AI-Platform-Tutorials/blob/master/docs/DPU-Integration/reference-files/files/face_detection/face_detection.cc

Python should work but you may need to find some reference design or write the code by yourself.

Best Regards,
Jason
-----------------------------------------------------------------------------------------------
Please mark the Answer as "Accept as solution" if the information provided is helpful.

Give Kudos to a post which you think is helpful and reply oriented.
-----------------------------------------------------------------------------------------------
0 Kudos