logo

Version 1.9.72 (Release Notes)


SOD Embedded Source Package Release 1.1.9 (July 2023, Changelog )

Open Source License
GPLv3 Logo

GPLv3 - Go Open Source

 Free of charge for open source projects or applications that are not distributed to third parties.
 Compatible with most open source licenses (GPLv3).
 Built-in RealNets Training Interfaces.
 Amalgamated, high code quality.
 Community Support.

  Download
Commercial License
PixLab | Symisc Systems

Commercial Licensing

 Multi-core CPU support for all platforms - Up to 3 ~ 10 times faster processing speed.
 Built-in (C Code), high performance RealNets frontal face detector.
 75 days of integration & technical assistance.
 Royalty-free commercial licenses without any GPL restrictions.
 Application source code stays private.
Notable SOD Users

New Model - Real-Time, WebAssemby, face detection model for Web apps. Find out more.


The sample set are practical usage, real-world working code implemented in C intended to familiarize the reader with the SOD Embedded API and is also available to consult online here. For an introduction course to the API, see Getting Started with SOD and The C/C++ API Reference Guide. You’re welcome to copy/paste and run these examples to see the API in action.

  Download Code Samples

Pre-trained CNN Models

Production ready, pre-trained models to be used in conjunction with the SOD CNN interfaces.

Model Total Classes Magic Word Model Size Description Usage Availability
CNN Face Detector 1 :face 396 KB Real-time, robust, multi-scale & shape invariant (i.e. frontal, inclined, large, tiny, etc.) face detection CNN model.
ID - face_cnn.sod
cnn_face.c $20 (One time fee)

Tiny Voc 20 :voc 60 MB Smallest & fastest object detection CNN model pre-trained on the Pascal VOC dataset that is able to detect 20 classes of different objects (i.e. car, person, dog, chair, etc.). This model works at Real-time on a Core I7 and similar CPUs with proper compiler optimizations (AVX, SSE, etc.) or using the proprietary multi-core enabled SOD release.
ID - tiny20.sod
cnn_voc.c $25 (One time fee)

Tiny COCO 80 :coco 61 MB Small & fast object detection CNN model pre-trained on the MS COCO dataset that is able to detect 80 classes of different objects (i.e. bus, person, airplane, stop sign, etc.). This model works at Real-time on a Core I7 and similar CPUs with proper compiler optimizations (AVX, SSE, etc.) or using the proprietary multi-core enabled SOD release.
ID - tiny80.sod
cnn_coco.c $20 (One time fee)

Full 80 :full 257MB Most accurate but largest & slowest (compared to :voc or :coco) object detection CNN model pre-trained on the MS COCO dataset that is able to detect 80 classes of different objects including car, motorbike, horse, bicycle, and so forth. This model does not work at Real-time even with the proprietary multi-core enabled SOD release.
ID - full.sod
cnn_full.c $29 (One time fee)

Sat 72 N/A 170MB Distill useful information including building, shapes, forms, etc. from satellite imagery at Real-time using the proprietary multi-core enabled SOD release. This model require prior approbation before delivery.
ID - sat.sod
N/A Request Model.

Pre-trained RealNets Models

Production ready, pre-trained models to be used in conjunction with the SOD RealNets interfaces.

Model Model Size RAM Consumption Description Usage Availability
Frontal Face Detector 234 KB < 10 MB Real-time (5 ~ 15 ms on HD video stream), frontal face detector RealNet model pre-trained on the Genki-4K datatset.
This is the recommended model if you are capturing video stream from user's Webcam or smartphone frontal camera to implement Snapchat-like filters, face recognition and so forth.
RealNets are designed to analyze & extract useful information from video stream rather than static images thanks to their fast processing speed (less than 10 milliseconds on 1920*1080 HD stream) and low memory footprint making them suitable for use on mobile devices. You are encouraged to connect the RealNets APIs with the OpenCV Video capture interfaces or any proprietary Video capture API to see them in action.
ID - face.realnet.sod
realnet_face.c $20 (One time fee)

WebAssemby Face Model 242 KB < 5 MB Frontal face detector, WebAssemby model pre-trained on the Genki-4K datatset for Web oriented applications.
The model is production ready, works at Real-Time on all modern browsers (mobile devices included). Usage instruction already included in the package.
ID - Webassembly.face.model
usage.html $20 (One time fee)


Pre-trained RNN Models

Production ready, pre-trained text generation models to be used in conjunction with the SOD RNN interfaces.

Model Magic Word Model Size RAM Consumption Description Gist Availability
Kant :rnn 34 MB < 35 MB RNN text generation model pre-trained on Immanuel Kant various essays.
ID - kant-rnn.sod
rnn_text.c $19 (One time fee)

4chan :rnn 34 MB < 35 MB RNN text generation model pre-trained on 4chan various Anon posts.
ID - anon_4chan.sod
rnn_text.c $19 (One time fee)

Tolstoy :rnn 34 MB < 35 MB RNN text generation model pre-trained on Leo Tolstoy various essays.
ID - tolstoy-rnn.sod
rnn_text.c $19 (One time fee)

Art History :rnn 34 MB < 35 MB RNN text generation model pre-trained on various art history essays.
ID - arthistory_rnn.sod
rnn_text.c $19 (One time fee)

Shakespeare :rnn 34 MB < 35 MB RNN text generation model pre-trained on William Shakespeare various essays.
ID - shk_rnn.sod
rnn_text.c $19 (One time fee)


Miscellaneous Models

Models not tied with the SOD library.

Model Model Size Description Web Demo Gist Availability
ASCII Art 2.6 MB Transform an input image or video frame into printable ASCII characters at real-time using a single decision tree. Real-time performance is achieved by using pixel intensity comparison inside internal nodes of the tree.
The library repository is available on Github.
This is the hex output model generated during the training phase. It contains both the codebook and the decision tree that let you render your images or video frames at Real-time.
ID - ascii_art.hex
art.pixlab.io sample.c $25 (One time fee)


Various Datasets

Useful public datasets.