Closed totovr closed 6 years ago
I do not think thr raspberry bsm gpu has enough RAM to support the model. You woukd have to use a re-trained simplified model, nit even sure that would work. Then you are running the performance question, you would likely ot even have 1fps on a rapi3. Tets
Actually you can not even use the rpi gpu here, no compute framework for deepnets available. So you will have to do 100% on CPU. Hardly conceivable, you will run short on RAM here as well anyway I think. With Rpi6 maybe...
You may want to close the issue here, not sure you will get much more I am afraid.
Tets
Thanks for your answer, it's the right one I am afraid it's not possible at the moment.
How about using Movidius NCS for CNN Inference in the Raspberry pi?
@crosofg have you achieved to use Movidius NCS with openpose?
We have used Open pose on NVIDIA Jetson TX2, it is a bit slow but working fine. Do anybody try to use it on another single board computer ? NVIDIA Jetson TX2 is quit expensive and not very affordable for IoT projects.
@crosofg @lciruelo OpenPose is not compatible with the movidius chip without some modification to the architecture. I have tried and it won't compile. The convolutional kernels must be <=5 and there are other restrictions, see here - https://aiyprojects.withgoogle.com/vision/#constraints
@ss32 @crosofg have you found a way to use Movidius NCS with openpose?
how about raspberry pi with Coral USB?
Hello Gines! I have a question, do you think that I could use openpose in a Raspberry 3 with Linux?