KSNN: OpenPose demo

Frank · October 16, 2021, 10:36am

Demonstration effect (On VIM3)

openpose

How to Run

Clone the github-directory:

$ git clone https://github.com/khadas/ksnn.git
$cd ksnn/examples/caffe

Plug in your webcam, and find it’s device number using the command:

ls /dev/video*

If you have multiple devices and are unsure which is your webcam, compare the command output between the unplugged and plugged-in states.

Single Person Demo

Replace ‘X’ with your webcam’s device number, e.g. 1

$ python3 openpose-signle-cap.py --model ./models/VIM3/openpose.nb --library libs/libnn_openpose.so --device X --level 0
$ python3 openpose-signle-picture.py --model ./models/VIM3/openpose.nb --library libs/libnn_openpose.so --picture data/person.jpg --level 0

Multiple Persons Demo

Replace ‘X’ with your webcam’s device number, e.g. 1

$ python3 openpose-multi-cap.py --model ./models/VIM3/openpose.nb --library libs/libnn_openpose.so --device X --level 0
$ python3 openpose-multi-picture.py --model ./models/VIM3/openpose.nb --library libs/libnn_openpose.so --picture data/person.jpg --level 0

Convert Parameters

$ ./convert \
--model-name openpose \
--platform caffe \
--model pose_deploy_linevec.prototxt \
--weights pose_iter_440000.caffemodel \
--mean-values '0,0,0,256' \
--quantized-dtype asymmetric_affine \
--kboard VIM3 --print-level 1

Model

The official original model needs to be modified before conversion, here is the modified model

github.com

yan-wyb/models-zoo/blob/master/caffe/openpose/pose_deploy_linevec.prototxt

#input: "image"
#input_dim: 1
#input_dim: 3
#input_dim: 1 # This value will be defined at runtime
#input_dim: 1 # This value will be defined at runtime
layer {
	name: "image"
	type: "Input"
	top: "image"
	input_param {
		shape {
			dim: 1
			dim: 3
			dim: 368
			dim: 368
		}
	}
}
layer {
  name: "conv1_1"

This file has been truncated. show original

https://gitlab.com/yan518/models-zoo-big/-/blob/master/openpose-368.caffemodel

Note

The model is the original model and needs to be optimized.
Maybe use AlphaPose instead.

jason_zhu · October 30, 2021, 7:46pm

Does KSNN use the 5.0 TOPS NPU?

Frank · October 31, 2021, 12:45am

@jason_zhu Yes. all demo and api use NPU

afa1414 · August 30, 2024, 10:11am

I found an old file and tried to run it, but the skeleton is not displayed on the webcam output. Is this a demo file that is no longer supported?