human-pose-estimation-0006

## Use Case and High-Level Description

This is a multi-person 2D pose estimation network based on the EfficientHRNet approach (that follows the Associative Embedding framework). 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 17 keypoints: ears, eyes, nose, shoulders, elbows, wrists, hips, knees, and ankles.

## Specification

Metric Value
Average Precision (AP) 51.1%
GFlops 8.844
MParams 8.1506
Source framework PyTorch*

Average Precision metric described in COCO Keypoint Evaluation site.

## Inputs

Name: input, shape: 1, 3, 352, 352. An input image 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 three blobs:

• heatmaps of shape B, 17, 176, 176 containing location heatmaps for keypoints of all types. Locations that are filtered out by non-maximum suppression algorithm have negated values assigned to them.
• embeddings of shape B, 17, 176, 176, 1 containing associative embedding values, which are used for grouping individual keypoints into poses.

## Legal Information

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