EDAboard.com | EDAboard.de | EDAboard.co.uk | WTWH Media

Neural Network on Xilinx Virtex 5

Ask a question - edaboard.com

elektroda.net NewsGroups Forum Index - VHDL Language - Neural Network on Xilinx Virtex 5

Goto page Previous  1, 2


Guest

Mon Feb 25, 2019 12:45 am   



On Sunday, February 24, 2019 at 4:09:14 PM UTC-5, Electronics_hobbyist wrote:
Quote:
On Sunday, February 24, 2019 at 9:11:34 PM UTC+2, gnuarm.del...@gmail.com wrote:
On Sunday, February 24, 2019 at 11:59:47 AM UTC-5, Electronics_hobbyist wrote:
On Sunday, February 24, 2019 at 4:56:02 PM UTC+2, gnuarm.del...@gmail..com wrote:
On Sunday, February 24, 2019 at 5:43:13 AM UTC-5, Electronics_hobbyist wrote:

I have the freedom to choose any solution I prefer. I am thinking of Convolution Neural Networks as I have read papers about this. But, my main concern is that on FPGAs there is a memory bottleneck. On the same FPGA I should use a core for the camera, the NN core and the serial interface core that will output me the result. Is there any source/link/book that could help me on how to start?

Before you do any writing of code, you should plan how you are going to debug it, with "it" being each the algorithm and the implementation.

If you are just connecting prewritten modules there shouldn't be much need to study books other than language books. Do you know which language you will be using? What language are the modules written in?

Rick C.

I have built it in python. But I will program it in VHDL for obvious reasons, i.e. faster decision of the NN if the person is man or woman...

Have you programmed in VHDL before?

Rick C.

Yes, many times, from simple digital design to FSMs. But first time I will program Neural Networks.


Ok, so the issues you face should be similar to the issues of using Neural Networks in python.

Of course, you will need to structure the algorithm to utilize parallelism to fully exploit the speed advantages of hardware.

Rick C.

Electronics_hobbyist
Guest

Mon Feb 25, 2019 10:45 pm   



On Monday, February 25, 2019 at 1:42:03 AM UTC+2, gnuarm.del...@gmail.com wrote:
Quote:
On Sunday, February 24, 2019 at 4:09:14 PM UTC-5, Electronics_hobbyist wrote:
On Sunday, February 24, 2019 at 9:11:34 PM UTC+2, gnuarm.del...@gmail.com wrote:
On Sunday, February 24, 2019 at 11:59:47 AM UTC-5, Electronics_hobbyist wrote:
On Sunday, February 24, 2019 at 4:56:02 PM UTC+2, gnuarm.del...@gmail.com wrote:
On Sunday, February 24, 2019 at 5:43:13 AM UTC-5, Electronics_hobbyist wrote:

I have the freedom to choose any solution I prefer. I am thinking of Convolution Neural Networks as I have read papers about this. But, my main concern is that on FPGAs there is a memory bottleneck. On the same FPGA I should use a core for the camera, the NN core and the serial interface core that will output me the result. Is there any source/link/book that could help me on how to start?

Before you do any writing of code, you should plan how you are going to debug it, with "it" being each the algorithm and the implementation.

If you are just connecting prewritten modules there shouldn't be much need to study books other than language books. Do you know which language you will be using? What language are the modules written in?

Rick C.

I have built it in python. But I will program it in VHDL for obvious reasons, i.e. faster decision of the NN if the person is man or woman...

Have you programmed in VHDL before?

Rick C.

Yes, many times, from simple digital design to FSMs. But first time I will program Neural Networks.

Ok, so the issues you face should be similar to the issues of using Neural Networks in python.

Of course, you will need to structure the algorithm to utilize parallelism to fully exploit the speed advantages of hardware.

Rick C.


Is there a helpful book/source I could start with?


Guest

Tue Feb 26, 2019 5:45 am   



On Monday, February 25, 2019 at 3:01:50 PM UTC-6, Electronics_hobbyist wrote:
Quote:
On Monday, February 25, 2019 at 1:42:03 AM UTC+2, gnuarm.del...@gmail.com wrote:
On Sunday, February 24, 2019 at 4:09:14 PM UTC-5, Electronics_hobbyist wrote:
On Sunday, February 24, 2019 at 9:11:34 PM UTC+2, gnuarm.del...@gmail..com wrote:
On Sunday, February 24, 2019 at 11:59:47 AM UTC-5, Electronics_hobbyist wrote:
On Sunday, February 24, 2019 at 4:56:02 PM UTC+2, gnuarm.del...@gmail.com wrote:
On Sunday, February 24, 2019 at 5:43:13 AM UTC-5, Electronics_hobbyist wrote:

I have the freedom to choose any solution I prefer. I am thinking of Convolution Neural Networks as I have read papers about this. But, my main concern is that on FPGAs there is a memory bottleneck. On the same FPGA I should use a core for the camera, the NN core and the serial interface core that will output me the result. Is there any source/link/book that could help me on how to start?

Before you do any writing of code, you should plan how you are going to debug it, with "it" being each the algorithm and the implementation.

If you are just connecting prewritten modules there shouldn't be much need to study books other than language books. Do you know which language you will be using? What language are the modules written in?

Rick C.

I have built it in python. But I will program it in VHDL for obvious reasons, i.e. faster decision of the NN if the person is man or woman....

Have you programmed in VHDL before?

Rick C.

Yes, many times, from simple digital design to FSMs. But first time I will program Neural Networks.

Ok, so the issues you face should be similar to the issues of using Neural Networks in python.

Of course, you will need to structure the algorithm to utilize parallelism to fully exploit the speed advantages of hardware.

Rick C.

Is there a helpful book/source I could start with?


What exactly do you want help with? What sort of book are you looking for?

Rick C.

Electronics_hobbyist
Guest

Tue Feb 26, 2019 10:45 am   



Sample code on VHDL, about Neural Networks, that is working. So I can start from somewhere to build something. From theory to practice there is a huge gap without help..

Thank you.


Guest

Tue Feb 26, 2019 3:45 pm   



On Tuesday, February 26, 2019 at 2:57:16 AM UTC-6, Electronics_hobbyist wrote:
> Sample code on VHDL, about Neural Networks, that is working. So I can start from somewhere to build something. From theory to practice there is a huge gap without help..

VHDL and Neural Networks are two totally different things. There is no need to combine them into one topic for study. If you have a working algorithm in another language, you don't really need to consider the Neural Networks so much. You just need to port that algorithm to VHDL like you would any algorithm. So I don't think you need or will find a book on this exact topic.

The things you will need to pay attention to are how you will adjust the algorithm to suit the inherent parallelism available in hardware and how you will debug the implementation, both in simulation and in the hardware.

Have you given this any thought? If you have questions you can ask here.

Rick C.

Electronics_hobbyist
Guest

Tue Feb 26, 2019 7:45 pm   



On Tuesday, February 26, 2019 at 4:39:38 PM UTC+2, gnuarm.del...@gmail.com wrote:
Quote:
On Tuesday, February 26, 2019 at 2:57:16 AM UTC-6, Electronics_hobbyist wrote:
Sample code on VHDL, about Neural Networks, that is working. So I can start from somewhere to build something. From theory to practice there is a huge gap without help..

VHDL and Neural Networks are two totally different things. There is no need to combine them into one topic for study. If you have a working algorithm in another language, you don't really need to consider the Neural Networks so much. You just need to port that algorithm to VHDL like you would any algorithm. So I don't think you need or will find a book on this exact topic.

The things you will need to pay attention to are how you will adjust the algorithm to suit the inherent parallelism available in hardware and how you will debug the implementation, both in simulation and in the hardware.

Have you given this any thought? If you have questions you can ask here.

Rick C.


Thanx a lot, for the help.


Guest

Tue Feb 26, 2019 9:45 pm   



On Tuesday, February 26, 2019 at 12:33:57 PM UTC-6, Electronics_hobbyist wrote:
Quote:
On Tuesday, February 26, 2019 at 4:39:38 PM UTC+2, gnuarm.del...@gmail.com wrote:
On Tuesday, February 26, 2019 at 2:57:16 AM UTC-6, Electronics_hobbyist wrote:
Sample code on VHDL, about Neural Networks, that is working. So I can start from somewhere to build something. From theory to practice there is a huge gap without help..

VHDL and Neural Networks are two totally different things. There is no need to combine them into one topic for study. If you have a working algorithm in another language, you don't really need to consider the Neural Networks so much. You just need to port that algorithm to VHDL like you would any algorithm. So I don't think you need or will find a book on this exact topic.

The things you will need to pay attention to are how you will adjust the algorithm to suit the inherent parallelism available in hardware and how you will debug the implementation, both in simulation and in the hardware.

Have you given this any thought? If you have questions you can ask here.

Rick C.

Thanx a lot, for the help.


What is the basis of the neural network algorithm? How is the software organized? I guess the key issue in performance are the high order loops, things that are iterated in nested loops. Have you thought much about how they will be implemented in VHDL?

Rick C.

Electronics_hobbyist
Guest

Wed Feb 27, 2019 11:45 am   



On Tuesday, February 26, 2019 at 10:12:59 PM UTC+2, gnuarm.del...@gmail.com wrote:
Quote:
On Tuesday, February 26, 2019 at 12:33:57 PM UTC-6, Electronics_hobbyist wrote:
On Tuesday, February 26, 2019 at 4:39:38 PM UTC+2, gnuarm.del...@gmail.com wrote:
On Tuesday, February 26, 2019 at 2:57:16 AM UTC-6, Electronics_hobbyist wrote:
Sample code on VHDL, about Neural Networks, that is working. So I can start from somewhere to build something. From theory to practice there is a huge gap without help..

VHDL and Neural Networks are two totally different things. There is no need to combine them into one topic for study. If you have a working algorithm in another language, you don't really need to consider the Neural Networks so much. You just need to port that algorithm to VHDL like you would any algorithm. So I don't think you need or will find a book on this exact topic.

The things you will need to pay attention to are how you will adjust the algorithm to suit the inherent parallelism available in hardware and how you will debug the implementation, both in simulation and in the hardware..

Have you given this any thought? If you have questions you can ask here.

Rick C.

Thanx a lot, for the help.

What is the basis of the neural network algorithm? How is the software organized? I guess the key issue in performance are the high order loops, things that are iterated in nested loops. Have you thought much about how they will be implemented in VHDL?

Rick C.


I have build the CNN in Raspberry in python but it takes 30-40 sec to decide if the photo shows a man or a woman. I want to accelarate this. How much faster can it go to the Virtex 5? Can I achieve 1-2 sec to take the decision?


Guest

Wed Feb 27, 2019 3:45 pm   



On Wednesday, February 27, 2019 at 3:52:27 AM UTC-6, Electronics_hobbyist wrote:
Quote:

I have build the CNN in Raspberry in python but it takes 30-40 sec to decide if the photo shows a man or a woman. I want to accelarate this. How much faster can it go to the Virtex 5? Can I achieve 1-2 sec to take the decision?


What you are asking is a bit like "how long is a piece of string?" I know nothing of the algorithm you implemented in python. The first step to speed this up is to identify where the processor is spending it's time. Have you done that?

To implement this algorithm in VHDL you will need to understand the algorithm enough that you can figure out how to implement different parts to run in parallel without one part still being a huge bottle neck. So you need to understand the timing of the code in python. Then you can figure out how to accelerate it. Just coding it in VHDL won't automatically make it run faster.

Can you explain the neural network code?

Rick C.

Goto page Previous  1, 2

elektroda.net NewsGroups Forum Index - VHDL Language - Neural Network on Xilinx Virtex 5

Ask a question - edaboard.com

Arabic version Bulgarian version Catalan version Czech version Danish version German version Greek version English version Spanish version Finnish version French version Hindi version Croatian version Indonesian version Italian version Hebrew version Japanese version Korean version Lithuanian version Latvian version Dutch version Norwegian version Polish version Portuguese version Romanian version Russian version Slovak version Slovenian version Serbian version Swedish version Tagalog version Ukrainian version Vietnamese version Chinese version Turkish version
EDAboard.com map