It looks like you're new here. If you want to get involved, click one of these buttons!Sign In
It looks like you're new here. If you want to get involved, click one of these buttons!
This will be a wandering statement from someone who is frustrated.
I am late to this Intel party? I read it was the Computer Vision SDK was now rebranded to OpenVino. What I didn't know is its
a deep learning thing too?
Guessing it came out around May. To be honest getting dizzy with all of whats going on.
From what I gather I can run this under Windows 10, thats way different.
And I can use the Movidius sticks? Does OpenCV use NVidia GPU acceleration or is that only tensorflow?
Does it magically use more than one Movidius stick? And is the new FPGA board for inference or creation?
This stuff is changing at a blistering pace.
To date, creating an input graph that works with Movidius has been an almost impossible challenge for me.
I have used and loved using the tools from Movidius but I can only modify and use the exmaple graphs.
I can not create my own custom input at all. Its not clear to me how its done since there are so many restrictions.
At least I have been reading that others have the same issues with CUSTOM graphs failing to compile.
I need to create new graphs for facial recognition (the example you have delivers awful results) and
image classification. I have created and retrained a tensorflow inceptionV3 graph that work using retain.py example
but I can not get it to compile. I posted it here with no response that I can recall.
I tried to do the webinar but never got an invite and I do not see a link to a video either. Last one I did glossed over the only think I care about
how to generate INPUT for this thing.
Now there is OpenVino whose example tosses in using Amazon services into the mix.
No idea how the Movidius sticks help with the inference but they are mentioned.
Can you sense the frustration as I am new to python, new to linux, and relatively new to Deep Learning, only working on and off with Tensorflow and Keras since last year.
Now there is OpenVino which goes back to Windows 10 where tools and most small IOT boards don't function.
Add to that entirely new examples of how to get this all to work.
By the way trying to multitask in python is exceptionally awful and I am glad to see OpenVino has addressed that by creating some sort of task scheduler.
Is there a question in there? Yeah which way should anyone go? And is there a cook book to create a custom graph you can actually compile that delivers good results.
Is there a way to accelerate OpenFace it seems to be much better at facial recognition.