lightweight-human-pose-estimation

Input

Ailia input shape: (1, 3, 240, 320)
Range: [-0.5, 0.5]

Output

Confidence: (1, 19, 30, 40)
Range: [0, 1.0]

Usage

Automatically downloads the onnx and prototxt files on the first run. It is necessary to be connected to the Internet while downloading.

For the sample image,

$ python3 lightweight-human-pose-estimation.py

If you want to specify the input image, put the image path after the --input option.
You can use --savepath option to change the name of the output file to save.

$ python3 lightweight-human-pose-estimation.py --input IMAGE_PATH --savepath SAVE_IMAGE_PATH

By adding the --video option, you can input the video.
If you pass 0 as an argument to VIDEO_PATH, you can use the webcam input instead of the video file.

$ python3 lightweight-human-pose-estimation.py --video VIDEO_PATH

The default setting is to use the optimized model and weights, but you can also switch to the normal model by using the --normal option.

Reference

Fast and accurate human pose estimation in PyTorch. Contains implementation of "Real-time 2D Multi-Person Pose Estimation on CPU: Lightweight OpenPose" paper.

Framework

Pytorch 1.2.0

Model Format

ONNX opset = 10

Netron

lightweight-human-pose-estimation.onnx.prototxt

lightweight-human-pose-estimation.opt.onnx.prototxt