ArducamIMX500SDK

Overview

---

Function	Support
Model input type	RGB, Y, YUV444, BGR, BAYER
Data injection image type	.jpg, .jpeg, .png, .bmp, .tif, .tiff, .webp, .ppm, .pgm, .pbm
Model config file	Config preprocess of input image by network_info.txt
Multi Model FPK	support
Model Size (quantified)	<= 8 MB
Model input data size (uint8)	<= 640 x 480 x 3
Task Converage	classification, object detection, pose estimation, segmentation

Install

Please visit the Releases page and download the corresponding .whl installation package based on your operating system and Python version.

After downloading, install it using the following command:

pip install your_package_name‑<version>‑<python_version>‑<platform>.whl

Example

Suppose you are using Windows and Python 3.8, and the downloaded file is:

arducamimx500sdk‑1.2.2‑cp38‑cp38‑win_amd64.whl

Then run the following command in the terminal:

pip install arducamimx500sdk‑1.2.2‑cp38‑cp38‑win_amd64.whl

Simple and Powerful API

ArducamIMX500SDK is designed to be both easy to use and highly extensible. Whether you're looking for a plug-and-play AI camera experience or planning to build complex computer vision pipelines, this SDK offers a clean and flexible Python interface.

🚀 Get Started

Clone this repository and run the following command in the project directory:

git submodule update --init --recursive

cd examples
# mobilenetv2
python app.py -pm mobilenetv2 -lf ..\firmware\arducam_imx500\loader.fpk -mf ..\firmware\arducam_imx500\firmware.fpk
# mobilenetssd
python app.py -pm mobilenetssd -lf ..\firmware\arducam_imx500\loader.fpk -mf ..\firmware\arducam_imx500\firmware.fpk
# yolov8n_det
python app.py -pm yolov8n_det -lf ..\firmware\arducam_imx500\loader.fpk -mf ..\firmware\arducam_imx500\firmware.fpk
# yolov8n-cls (yolov8n_cls is also accepted)
python app.py -pm yolov8n-cls -lf ..\firmware\arducam_imx500\loader.fpk -mf ..\firmware\arducam_imx500\firmware.fpk
# yolov8n-seg (yolov8n_seg is also accepted)
python app.py -pm yolov8n-seg -lf ..\firmware\arducam_imx500\loader.fpk -mf ..\firmware\arducam_imx500\firmware.fpk
# yolov8n_pos
python app.py -pm yolov8n_pos -lf ..\firmware\arducam_imx500\loader.fpk -mf ..\firmware\arducam_imx500\firmware.fpk
# yolov8n_pos_hand
python app.py -pm yolov8n_pos_hand -lf ..\firmware\arducam_imx500\loader.fpk -mf ..\firmware\arducam_imx500\firmware.fpk
# deeplabv3plus
python app.py -pm deeplabv3plus -lf ..\firmware\arducam_imx500\loader.fpk -mf ..\firmware\arducam_imx500\firmware.fpk

Sample App

git submodule update --init --recursive

cd examples
# geofencing
python app.py -dp geofencing -lf ..\firmware\arducam_imx500\loader.fpk -mf ..\firmware\arducam_imx500\firmware.fpk
# detect and count package
python app.py -dp package -lf ..\firmware\arducam_imx500\loader.fpk -mf ..\firmware\arducam_imx500\firmware.fpk

Command-line Options

usage: app.py [-h] [-wf] [-lf LOADER_FIRMWARE] [-mf MAIN_FIRMWARE] [-m MODEL] [-dp DEMO_PROJECT] [-pm PRETRAIN_MODEL]
              [-d DEVICE_ID] [-dy] [-dyr] [-di DATA_INJECTION] [--network-info NETWORK_INFO]
              [--rect-crop XMIN YMIN XMAX YMAX] [--fps FPS]

options:
  -h, --help            show this help message and exit
  -wf, --write-flash    Flag of flash write.
  -lf, --loader-firmware LOADER_FIRMWARE
                        Loader firmware path.
  -mf, --main-firmware MAIN_FIRMWARE
                        Main firmware path.
  -m, --model MODEL     Model path.
  -dp, --demo-project DEMO_PROJECT
                        Demo project name.
  -pm, --pretrain-model PRETRAIN_MODEL
                        Pretrain model name.
  -d, --device-id DEVICE_ID
                        Device Index. (default: 0)
  -dy, --dump-yuv       Dump YUV.
  -dyr, --dump-yuv-raw  Dump raw YUV.
  -di, --data-injection DATA_INJECTION
                        Data injection.
  --network-info NETWORK_INFO
                        network_info.txt path
  --rect-crop XMIN YMIN XMAX YMAX
                        Rect crop area in absolute xyxy format. X range: 0-4056, Y range: 0-3040.
  --fps FPS             Framerate

Data Injection Export Example

examples/data_injection_export.py is a one-shot data injection helper for offline export.

cd examples
# Use pretrain model
python data_injection_export.py -pm mobilenetssd -lf ..\firmware\arducam_imx500\loader.fpk -mf ..\firmware\arducam_imx500\firmware.fpk -i ..\pics\test.jpg -o .\outputs --fps 20

# Or use custom model path
python data_injection_export.py -m ..\model\arducam_imx500_model_zoo\mobilenetssd\network.fpk -lf ..\firmware\arducam_imx500\loader.fpk -mf ..\firmware\arducam_imx500\firmware.fpk -i ..\pics\test.jpg -o .\outputs --fps 20

# Batch mode: -i can be a directory
python data_injection_export.py -pm mobilenetssd -lf ..\firmware\arducam_imx500\loader.fpk -mf ..\firmware\arducam_imx500\firmware.fpk -i ..\pics\test -o .\outputs --fps 20

Outputs are generated in -o/<input_image_stem>/ and include:

• image.png
• input_tensor.png
• parsed_metadata.json

Ready for Advanced Development

If you plan to build your own applications, pipelines, or integrate with other systems, these APIs give you the control and flexibility you need.

📚 For full API reference and data structures, please refer to the official documentation: ArducamIMX500SDK.pdf

This document includes detailed explanations of class methods, parameter definitions, and usage examples — essential for anyone doing secondary development on the Arducam IMX500 platform.