MediaPipe Palm Detection

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

0 Likes

Model Details

Model Description

The MediaPipe Palm detection model is a single-shot detector optimized for mobile realtime applications. The model detects palms instead of entire hands since estimating bounding boxes of rigid objects like palms and fists is significantly simpler than detecting hands with articulated fingers.

Developed by: Google
Shared by:
Model type: Computer Vision
License:
Resources for more information:

Training Details

Training Data

For training the palm detector, in-the-wild dataset was used, which is sufficient for localizing hands and offers the highest variety in appearance.

In-the-wild dataset contains 6K images of a large variety, e.g. geographical diversity, various lighting conditions, and hand appearance. The limitation of this dataset is that it doesn’t contain complex articulation of hands.

Testing Details

Metrics

Evaluation results are taken from .

Metric	Value
Average Precision	95.7%

Technical Specifications

Input/Output Details

Input:
- Name: input_1
  - Info: NCHW BGR un-normalized image
Output:
- Name: multiple (see NN archive)
  - Info: Hand bounding boxes and classification scores

Model Architecture

BlazePalm single-shot detector and encoder-decoder feature extractor similar to FPN.

Please consult for more information on model architecture.

Throughput

Model variant: mediapipe-palm-detection:192x192

• Input shape: [1, 3, 192, 192] • Output shapes: [[1, 2016, 18], [1, 2016, 1]]

• Params (M): 1.133 • GFLOPs: 0.354

Platform	Precision	Throughput (infs/sec)	Power Consumption (W)
RVC2	FP16	51.18	N/A
RVC4	INT8	621.46	3.01

Model variant: mediapipe-palm-detection:128x128

Platform	Precision	Throughput (infs/sec)	Power Consumption (W)
RVC2	FP16	69.41	N/A

* 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 30 images from the web where the hands of the people are present.

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/mediapipe-palm-detection:192x192"
)

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

Inspect model head(s):

MPPalmDetectionParser that outputs message (bounding boxes of detected hands with confidence scores).

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/mediapipe-palm-detection:192x192

Moreover, you can also check and examples.

Palm detection model.
License	Apache 2.0 Commercial use
Downloads	693
Tasks	Object Detection
Model Types	IR ONNX

Model Variants

Name	Version	Available For	Created At	Deploy
		RVC2	11 months ago
		RVC2, RVC4	11 months ago