STMicroelectronics/head_landmarks

Head landmarks quantized

Use case : Pose estimation

Model description

Head landmarks is a single pose estimation model targeted for real-time processing implemented in ONNX.

The model is quantized in int8 format using onnx quantizer.

Network information

Network information	Value
Framework	ONNX Runtime
Quantization	int8
Provenance	https://github.com/PINTO0309/PINTO_model_zoo/tree/main/032_FaceMesh
Paper	https://developers.google.com/ml-kit/vision/face-mesh-detection

Networks inputs / outputs

With an image resolution of NxM with K keypoints to detect :

Input Shape	Description
(1, N, M, 3)	Single NxM RGB image with UINT8 values between 0 and 255

Output Shape	Description
(1, 1, 1, Kx2)	FLOAT values Where Kx2 are the (x,y) values of each keypoints

Recommended Platforms

Platform	Supported	Recommended
STM32L0	[]	[]
STM32L4	[]	[]
STM32U5	[]	[]
STM32H7	[]	[]
STM32MP1	[x]	[]
STM32MP2	[x]	[x]
STM32N6	[x]	[x]

Performances

Metrics

Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.

Reference NPU memory footprint

Model	Format	Resolution	Series	Internal RAM (KiB)	External RAM (KiB)	Weights Flash (KiB)	STM32Cube.AI version	STEdgeAI Core version
head_landmarks	Int8	224x224x3	STM32N6	1739.5	0.0	3246.47	10.2.0	2.2.0

Reference NPU inference time

Model	Format	Resolution	Board	Execution Engine	Inference time (ms)	Inf / sec	STM32Cube.AI version	STEdgeAI Core version
head_landmarks	Int8	224x224x3	STM32N6570-DK	NPU/MCU	20.52	48.73	10.2.0	2.2.0