facial-landmarks-98-detection-0001

Use Case and High-Level Description

This is a 2D facial landmarks detection network based on the HRNet approach. For face in an image, the network detects landmarks (look at image below). The landmarks contain 98 keypoints.

Dataset (training and validation) - Internal

Network is trained and validated on the custom dataset based on WiderFace and VGG2 subsets.

Example

Specification

Metric

Value

NME

0.1323

GFlops

0.6

MParams

9.66

Source framework

PyTorch*

Inputs

Name: input.1, shape: 1, 3, 64, 64. 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 a blob 3851 with the shape: 1, 98, 16, 16, containing location heatmaps for 98 keypoints. Locations that are filtered out by non-maximum suppression algorithm have negated values assigned to them.

Demo usage

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