Winners Announcement: PYNQ-Z2: Embedded Vision Workshop (Zero to Hero) Series with Adam Taylor!

to add on, initially i am also confuse how to enrol, after you clcik to individual session, you see list of attendees, you click the somewhere above there to enrol (there is no obvious 'button' but the link is working

after further checking, the ultra96 is running yolo off PL (programmable logic) fabric for the computational expensive convolution calculation. Although ultra96 is not pynq-z2, both are using pynq, i not sure if that means pynq-Z2 can run yolo example by just copy over the demo or we need to do some tinkering with overlay. based on my current understanding (refer to the pynq diagram below), to make use of PL for acceleration, it need a platform specific overlay.

i watched a webinar, and got helpful reply from bhfletcher here

https://www.element14.com/community/events/5519/l/using-python-overlays-to-experiment-with-neural-networks#comment-25572…

I am working on a startup idea in the logistics and freight carrying sector, which originated when observing the market potential related to the struggles my parents' trucking company had over the last decade.

The product consists of a combination of HaaS and SaaS offering. The HaaS category is my current focus and is the one I have made the most progress in so far; it consists of:

• a sensor mesh network;
• a bidirectional CAN to wireless bridge device for communication with the vehicle;
• a GPS/Cellular gateway with LoRa subsystem for a custom VANET implementation (topic of my Master Thesis) for distributed peer to peer communication between vehicles and routing of packets without the use of commercial LTE networks;
• an FPGA based Embedded Vision SBC for edge case detection and CCTV footage collection/processing.

The content of these webinars will allow me to gain an edge in implementing these algorithms for the latter aspect of my project and allow me to take the project idea to a complete product as I had envisioned it. I am lucky to already have access to a Pynq Z2 board, thanks to our course of Digital Systems at KU Leuven, but I will have to return this board before the 2nd webinar takes place, leaving me without the resources to follow along the webinars and hence not allowing me to implement the part of the product I mentioned above for some time (until I am able to purchase a similar device) despite of this great webinar initiative purpose. I would be grateful if I was nominated for one of the PYNQ Z2 boards in order not to have this delay/inability to realize the idea due to temporary inability to purchase own device.

I am extremely excited to dive into the webinars and have already began reading the MPSoC book in between completing the university assignments! Regardless, I will attempt to get the most out of these webinars with the time I have before having to return the current board back to the university!

"...Presently, I do not see an enroll button.

How may I enrol?..."

You need to select the individual workshop session pages and enrol for each session there.

PYNQ-Z2 Workshop: Getting Started with PYNQ

PYNQ-Z2 Workshop: Getting Up and Running with PYNQ

PYNQ-Z2 Workshop: Unlocking Your Inner PYNQ Hero

Hi,

I do not need the board, as I have already got one from Element14 as part of a previous road test.

However, I would like to enroll to learn more about how to design overlays with Xilinx SDK like SDSoC, Vivado etc. (whatever is on offer).

I feel this is one area that needs really long and in depth tutorials since FPGA development has moved a long way further from VHDL/ Verilog to C/C++

Presently, I do not see an enroll button.

How may I enroll?

Best Regards
Sambit

I'm looking to make a prototype of a single board computer using a RISC-V CPU in the SoC, but there aren't any reasonably priced RISC-V based FPGA systems available for purchase yet. This board and its implemented framework seem like a great starting point to understanding FPGA programming, and trying to compile OSes in ARM seems like a less daunting task to accomplish. I was also thinking of how I could implement this framework for my own project.

I also have an idea for an easy to use all in one recording device.

There's just so many things I want to do I need to see what I can even pull off first before deciding on a direct path.

interesting...i once have tried to use zynq to emulate my tv ir remote control. didn't complete it as back then there is no pynq and i was using zc702, and the relative complexity of doing it with non high level language was a killer

i feel pynq will make this possible.

this is what i mean with android + roku as a mobile hdmi output camera

thanks for the link, i did read through this project but as it has the Intel Movidius it gives me the impression that it is offloading the yolo to the movidius rather than running through the FPGA fabric. I might be wrong, but i do try to relook at the link you posted and go more details into its shared project file. Based on https://github.com/fpgadeveloper/pynq-ncs-yolo/blob/master/notebooks/yolo-hdmi.ipynb  , it is using HDMI (so it pretty near to what this workshop is gonna be using), but after that setting up for movidius and going through movidius. It also doesn't seem to be using overlay for image processing, although i might be wrong.

Anyhow, i did continue my exploration, and have just watched the webinar Using Python Overlays to Experiment with Neural Networks organized by tariq.ahmad recently, and find some useful reference. During the webinar presented by bhfletcher, some built in example of the ultra96 MPSoC board is shown, specifically it is using overlay. BNN vgg16, QNN with image net classifier is showcased. I am pretty amaze that may be the QNN with imageNet is sufficient for my use case as many of the animals have being pretrained. One can browse the training set here, here and here. the 3rd here is interesting as i read on the comment that 2nd here doesn't match the labels used to train keras inception v3. But the good thing for me is that i found guinea pig is a trained classified animal in the image net. Likewise, cat is classfied too. So with imagenet alone probably is suitable for my application. However, image net is not an RCNN, meaning although it can classify, i believe it can't give the bounding box. So i am still eager to see whether yolo can be used as with the bounding we can know where captain hook is and shoot the water jet towards the direction.

figure cat on imagenet

Most current ML projects are limited by the processing rate of the system. Realistically, if the cameras an driver are not in sync, you get blurs for moving objects. LPC systems would have two parts. 1. The capture and image conditioning system and 2. the actual ML system.

Just my two cents.