cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
qwqw0001
Contributor
Contributor
570 Views
Registered: ‎07-15-2019

Can I use DNNDK for my model

Jump to solution

hi

  I want use Alexnet, but Xilinx doesn't provide a demo.So,can I use DNNDK for my model? And if I want to use the VGG example provided by Xilinx, can I change its network,?For example, delete its full connection layer and not do padding.

  Any advice is helpful

 ylh

Tags (1)
0 Kudos
1 Solution

Accepted Solutions
jasonwu
Moderator
Moderator
513 Views
Registered: ‎03-27-2013

Hi @qwqw0001 ,

 

For deploying a Convent on FPGA using DNNDK the pb file(trained model) should be provided by end customer.

Then you can use DNNDK to compress and prune the model.

You can refer to DNNDK tutorial to get more details:

https://www.xilinx.com/support/documentation/sw_manuals/ai_inference/v1_6/ug1327-dnndk-user-guide.pdf

 

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
6 Replies
jasonwu
Moderator
Moderator
555 Views
Registered: ‎03-27-2013

Hi @qwqw0001 ,

 

In my opinion if you use DPU/DNNDK supported layers to construct Convnet it should be supported.

Here are 2 simple examples about how to train custom model and deploy it on MPSoC devices with DNNDK 3.1(Xilinx tools v2019.1):

https://github.com/gewuek/flower_classification_dnndk_v1

https://github.com/gewuek/flower_classification_dnndk_v2

Hope these can help.

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
qwqw0001
Contributor
Contributor
541 Views
Registered: ‎07-15-2019

hi @jasonwu ,

 Thanks for your reply. And I was also wondering if I could make changes to the VGG model provided by Xilinx, such as removing the full connection layer, after all, it is pruned and compressed by Xilinx, and pruning is not free for me.

  ylh

0 Kudos
jasonwu
Moderator
Moderator
538 Views
Registered: ‎03-27-2013

Hi @qwqw0001 ,

 

When you mentioned "VGG model provided by Xilinx" which one do you mean?

And BTW as I know most of the demo models we  provided are just pb files. If you want to modify the network you need to train the model by yourself. That means you need to do that from training to deployment all along.

Please feel free to comment if I have missed anything here.

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
qwqw0001
Contributor
Contributor
530 Views
Registered: ‎07-15-2019

hi @jasonwu ,

   Thanks for your reply. In the DPU documentation :" It includes a set of highly optimized instructions, and supports most convolutional neural networks, such as VGG,ResNet, GoogLeNet, YOLO, SSD, MobileNet, FPN,and others."

    As you said, if only pb files are provided, is the network structure of pb files compressed and pruned?

    ylh

0 Kudos
jasonwu
Moderator
Moderator
514 Views
Registered: ‎03-27-2013

Hi @qwqw0001 ,

 

For deploying a Convent on FPGA using DNNDK the pb file(trained model) should be provided by end customer.

Then you can use DNNDK to compress and prune the model.

You can refer to DNNDK tutorial to get more details:

https://www.xilinx.com/support/documentation/sw_manuals/ai_inference/v1_6/ug1327-dnndk-user-guide.pdf

 

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
qwqw0001
Contributor
Contributor
509 Views
Registered: ‎07-15-2019
I get it,thank you!
0 Kudos