Running Video Capture Demo on Windows PC

[Greendogo]

Hey there, could you provide some instruction on setting up and running inference using a webcam on a Windows PC?

I'm getting stuck at the 'tvm' part.

[sushil-bharati]

Yes, this would be very helpful for testing the algorithm on various devices running Windows. It would also be better if you could give more information on how one could enable/disable GPU device(s). Thanks

[kevkid]

Same here, using the jetson demo, it states ModuleNotFoundError: No module named 'tvm'

[lmxyy]

The nano_demo is tested on Jetson Nano with TVM support. If you are using Jetson Nano, you could follow this guide to install TVM. If you are using other devices, @MemorySlices could you adapt the TVM demo to a PyTorch model one for a more general demo?

[sushil-bharati]

@lmxyy Do you know if the models are CPU friendly? Do we "require" GPU to run them optimally? I tried it in my CPU-only environment and it takes ~1.96 sec to process a frame (448x448x3). Am I doing sth wrong?

[lmxyy]

The model should be CPU-friendly, as we also include some results of Raspberry Pi and it only takes ~100ms. But if you directly run the PyTorch model using CPU, I think your result is reasonable, as the CPU backend is not well-optimized.

[sushil-bharati]

Thank you, @lmxyy for the prompt response. That explains why I am getting such a slow speed. I am indeed using model(s) using Pytorch's CPU backend settings. So, is there a way that I can run the optimized model(s) on a CPU-only env, or is that out of scope?

[lmxyy]

You could try TVM to optimize your CPU backend. But I think this will cost your much more time...

[kevkid]

Hi @sushil-bharati would it be possible to share how you got it to run using the pytorch cpu backend? I tried doing model(img) and got:

conv2d() received an invalid combination of arguments - got (numpy.ndarray, Parameter, NoneType, tuple, tuple, tuple, int), but expected one of:
 * (Tensor input, Tensor weight, Tensor bias, tuple of ints stride, tuple of ints padding, tuple of ints dilation, int groups)
      didn't match because some of the arguments have invalid types: (numpy.ndarray, Parameter, NoneType, tuple, tuple, tuple, int)
 * (Tensor input, Tensor weight, Tensor bias, tuple of ints stride, str padding, tuple of ints dilation, int groups)
      didn't match because some of the arguments have invalid types: (numpy.ndarray, Parameter, NoneType, tuple, tuple, tuple, int)

Thank you

[731076467]

Hello, I'd like to ask why I can't find this scheduler in the sentence from scheduler import warmup designer in dist_train file. What's the reason?

[MemorySlices]

Hi, Please ignore it and delete the corresponding import.

Best, Yihan

731076467 @.***> 于 2022年8月2日周二 上午11:03写道：

Hello, I'd like to ask why I can't find this scheduler in the sentence from scheduler import warmup designer in dist_train file. What's the reason?

— Reply to this email directly, view it on GitHub https://github.com/mit-han-lab/litepose/issues/13#issuecomment-1201961015, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANUHL3R34OKYCFMODBK7RCLVXCFX3ANCNFSM54FEHHLA . You are receiving this because you were mentioned.Message ID: @.***>