human-pose-estimation-0001

February 10, 2022 ยท View on GitHub

Use Case and High-Level Description

This is a multi-person 2D pose estimation network (based on the OpenPose approach) with tuned MobileNet v1 as a feature extractor. For every person in an image, the network detects a human pose: a body skeleton consisting of keypoints and connections between them. The pose may contain up to 18 keypoints: ears, eyes, nose, neck, shoulders, elbows, wrists, hips, knees, and ankles.

Example

Specification

MetricValue
Average Precision (AP)42.8%
GFlops15.435
MParams4.099
Source frameworkCaffe*

Average Precision metric described in COCO Keypoint Evaluation site.

Tested on a COCO validation subset from the original paper Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields.

Inputs

Image, name: data, shape: 1, 3, 256, 456 in the B, C, H, W format, where:

  • B - batch size
  • C - number of channels
  • H - image height
  • W - image width

Expected color order is BGR.

Outputs

The net outputs are two blobs:

  1. Name: Mconv7_stage2_L1, shape: 1, 38, 32, 57 contains keypoint pairwise relations (part affinity fields).
  2. Name: Mconv7_stage2_L2, shape: 1, 19, 32, 57 contains keypoint heatmaps.

Demo usage

The model can be used in the following demos provided by the Open Model Zoo to show its capabilities:

[*] Other names and brands may be claimed as the property of others.