BlazeFace Front 128x128 Quantized

April 16, 2026 ยท View on GitHub

Use case : Object detection

Model description

BlazeFace Front 128x128 is a lightweight and efficient face detection model optimized for real-time applications on embedded devices. It is a variant of the BlazeFace architecture, designed specifically for detecting frontal faces at a resolution of 128x128 pixels. The model is quantized to int8 format using TensorFlow Lite converter to reduce memory footprint and improve inference speed on resource-constrained hardware.

BlazeFace is known for its fast inference and accuracy, making it suitable for applications such as face tracking, augmented reality, and user authentication.

Network information

Network informationValue
FrameworkTensorFlow Lite
Quantizationint8
Input resolution128x128
Provenancehttps://github.com/PINTO0309/PINTO_model_zoo/tree/main/030_BlazeFace

Network inputs / outputs

Input ShapeDescription
(1, 128, 128, 3)Single 128x128 RGB image with FLOAT32 values between -1 and 1
Output ShapeDescription
(1, 512, 16)FLOAT32 tensor containing bounding box coordinates and keypointss for detected faces
(1, 512, 1)FLOAT32 tensor containing confidence scores for detected faces
(1, 384, 16)FLOAT32 tensor containing bounding box coordinates and keypointss for detected faces
(1, 384, 1)FLOAT32 tensor containing confidence scores for detected faces
PlatformSupportedRecommended
STM32L0[][]
STM32L4[][]
STM32U5[][]
STM32H7[][]
STM32MP1[][]
STM32MP2[][]
STM32N6[x][x]

Performances

Metrics

Performance metrics are measured using default STM32Cube.AI configurations with input/output allocated buffers.

ModelDatasetFormatResolutionSeriesInternal RAM (KB)External RAM (KB)Weights Flash (KB)STEdgeAI Core version
BlazeFace Front 128x128 per channelWIDER FACE (frontal)Int8128x128x3STM32N65280150.974.0.0

Reference NPU inference time (example)

ModelDatasetFormatResolutionBoardExecution EngineInference time (ms)Inf / secSTEdgeAI Core version
BlazeFace Front 128x128 per channelWIDER FACE (frontal)Int8128x128x3STM32N6570-DKNPU/MCU5.09196.34.0.0

Integration and support

For integration examples and additional services, please refer to the STM32 AI model zoo services repository:
https://github.com/STMicroelectronics/stm32ai-modelzoo-services

References