YuNet | Luxonis

Our new model ZOO works with DepthAI V3. Find out more in our documentation.

1+ Likes

Model Details

Model Description

YuNet is a highly efficient face detection model. With only 75k parameters, it is designed to operate with minimal computational resources while maintaining exceptional balance between accuracy and speed, making it ideal for real-time applications on edge devices.

Developed by: Wei Wu et al.
Shared by:
Model type: Computer Vision
License:
Resources for more information:

Photo by Sagar Sintan from

Training Details

Training Data

The model was trained on . It is split into training, validation and test sets. WIDERFace is a face detection benchmark dataset, of which images are selected from the publicly available . They choose 32,203 images and label 393,703 faces with a high degree of variability in scale, pose, and occlusion as depicted in the sample images. WIDERFace dataset is organized based on 61 event classes. For each event class, they randomly select 40%/10%/50% data as training, validation, and testing sets. Based on the detection rate of EdgeBox (Zitnick & Dolla ́r, 2014), three levels of difficulty (i.e. Easy, Medium and Hard) are defined by incrementally incorporating hard samples.

Testing Details

Metrics

AP metric was calculated for different groups of validation dataset based on the face-detection difficulty level (Easy, Medium, and Hard). The results are taken from the repository.

Metric	Value
AP_easy	0.887
AP_medium	0.871
AP_hard	0.768

Technical Specifications

Input/Output Details

Input:
- Name: image
  - Info: NCHW BGR image
Output:
- Name: Multiple (please consult NN archive config.json)
  - Info: Classification scores, objectness scores, bounding boxes, and keypoints for a multitude of detections.

Model Architecture

Backbone: Compact feature extraction backbone originating from MobileNet
Neck: Simplified feature pyramid network (FPN) neck
Head: Anchor-free head

See the for more information on model architecture.

Throughput

Model variant: yunet:960x720

• Input shape: [1, 3, 720, 960] • Output shapes: [[39615, 14], [39615, 2], [39615, 1]]

• Params (M): 0.083 • GFLOPs: 0.943

Platform	Precision	Throughput (infs/sec)	Power Consumption (W)
RVC2	FP16	16.66	N/A
RVC4	INT8	330.15	3.45

Model variant: yunet:640x480

• Input shape: [1, 3, 480, 640] • Output shapes: [[17610, 14], [17610, 2], [17610, 1]]

• Params (M): 0.083 • GFLOPs: 0.419

Platform	Precision	Throughput (infs/sec)	Power Consumption (W)
RVC2	FP16	38.61	N/A
RVC4	INT8	705.71	3.14

Model variant: yunet:1280x960

• Input shape: [1, 3, 960, 1280] • Output shapes: [[70500, 14], [70500, 2], [70500, 1]]

• Params (M): 0.083 • GFLOPs: 1.678

Platform	Precision	Throughput (infs/sec)	Power Consumption (W)
RVC4	INT8	178.93	3.35

Model variant: yunet:320x240

• Input shape: [1, 3, 240, 320] • Output shapes: [[4385, 14], [4385, 2], [4385, 1]]

• Params (M): 0.083 • GFLOPs: 0.105

Platform	Precision	Throughput (infs/sec)	Power Consumption (W)
RVC2	FP16	143.56	N/A
RVC4	INT8	694.17	2.46

* Benchmarked with , using 2 threads (and the DSP runtime in balanced mode for RVC4).

* Parameters and FLOPs are obtained from the package.

Quantization

RVC4 version of the model was quantized using a custom dataset. This was created by taking a 50-image subset of the dataset.

Utilization

Models converted for RVC Platforms can be used for inference on OAK devices. DepthAI pipelines are used to define the information flow linking the device, inference model, and the output parser (as defined in model head(s)). Below, we present the most crucial utilization steps for the particular model. Please consult the docs for more information.

Install DAIv3 and depthai-nodes libraries:

pip install depthai
pip install depthai-nodes

Define model:

model_description = dai.NNModelDescription(
    "luxonis/yunet:640x480"
)

nn = pipeline.create(ParsingNeuralNetwork).build(
    <CameraNode>, model_description
)

Inspect model head(s):

YuNetParser that outputs message (bounding boxes with keypoints and confidence scores for every detected face).

Get parsed output(s):

while pipeline.isRuning():
    parser_output: ImgDetectionsExtended = parser_output_queue.get()

Example

You can quickly run the model using our script. It automatically downloads the model, creates a DepthAI pipeline, runs the inference, and displays the results using our DepthAI visualizer tool. To try it out, run:

python3 main.py \
    --model luxonis/yunet:640x480

You can also check out and or any other example from .

Face detection model.
License	MIT Commercial use
Downloads	7864
Tasks	Object Detection
Model Types	ONNX

Model Variants

Name	Version	Available For	Created At	Deploy
		RVC4	9 months ago
		RVC2, RVC4	10 months ago
		RVC2, RVC4	10 months ago
		RVC2, RVC4	10 months ago