cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Highlighted
Newbie
Newbie
1,126 Views
Registered: ‎11-14-2018

suggestion for Object Detection on Zynq hardware: initial stage

Hello,

 
I am starting a project for object detection and recognition on my Zynq hardware. I came across several libraries from Xilinx (like CHaiDNN, etc.). There are many trained networks available like AlexNet, VGG, SSD, etc. My question is: does it also support transfer learning? Can I use one of the available pre-trained network like SSD, train it with a few images of faces and expect the model to have learnt to detect faces? I want to know if CHaiDNN supports that.  
 
I am also aware that a model trained on Caffe can be ported on Zynq hardware. 
 
Additionally, if anyone here is aware of TensorFlow, I currently use the Object Detection API and customize the model to detect my object of interest (example: faces). Now this is what I’m looking to do on Zynq. 
 
Which workflow would you recommend? Is it easier for me to train a model on Caffe and port it on my Zynq hardware or should I use something like CHaiDNN? 
 
If it helps: my trained model should work on Zynq based hardware, bare-metal, and the task would be to detect only one or two objects (cat / face / hand)
 
 
Tags (1)
0 Kudos
3 Replies
Highlighted
Explorer
Explorer
1,097 Views
Registered: ‎06-09-2015

On Tensorflow on Zynq:
Here is a tutorial for "Installing Tensorflow on PYNQ FPGA". Link: Hackster IO

Regards,
krishna@logictronix.com
0 Kudos
Highlighted
Moderator
Moderator
1,084 Views
Registered: ‎08-20-2018

Hi @meghanadk

It will be easier to use your SSD caffe model in CHaiDNN, as CHaiDNN library provides you APIs for processing, read/write, execution.  

 

Best Regards,
Nutan
-------------------------------------------------------------------------------
Please don't forget to reply, kudo and accept as a solution
Highlighted
Newbie
Newbie
1,065 Views
Registered: ‎11-14-2018

@nutang and @krishnagaihre, thanks for your response.
Please confirm if this model can be trained again using images with object of interest (example: faces) (Transfer learning) if I use CHaiDNN with caffe.
If yes, please direct me to useful resources which shows how I can re-use any of the CHaiDNN pre-trained models.
0 Kudos