Skip to content

About

Content for Neat workshop

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Building Edge AI Video Apps with SiMa Neat on Modalix

A hands-on workshop (about 2.5 hours) for building real-time video AI apps in C++ with the SiMa Neat Library on a Modalix DevKit.

You will:

  1. Compile two YOLO26 models yourself, one in INT8 and one in BF16.
  2. Build a single-stream app: one video source → one model → live video with boxes in Neat Insight.
  3. Build a dual-stream app: two video sources → two models (detection and pose) → two live outputs.

Tested with: Modalix DevKit, board software 2.1.3, Neat SDK 2.1.3.1, Neat Library 0.5.0.

Start here

Open 00-agenda.md. It has the schedule, the installation checklist, and links to every session.

# Session Guide Code
0 Installation and introduction 00-agenda.md
1 Neat C++ API introduction 01-neat-cpp-api.md model-info/
2 Model compilation: INT8 and BF16 02-model-compilation.md model-compilation/
3 App 1: one source, one model, one stream 03-app1-single-stream.md app1-single-stream/
4 App 2: two sources, two models, two streams 04-app2-dual-stream.md app2-dual-stream/

Getting the material

The guides expect the folder at /workspace/workshop inside the SDK container:

cd /workspace
git clone https://github.com/smaurya01/neat-workshop.git workshop

Prebuilt models

If you can't compile the models yourself, ask the instructor for the prebuilt model archives (*.tar.gz). They are not stored in this repository. Put them in a prebuilt/ folder, then see Prebuilt fallback for where to copy them.

Repository layout

├── 00-agenda.md … 04-app2-dual-stream.md   workshop guides
├── model-info/            session 1: inspect and benchmark a model
├── model-compilation/     session 2: export_onnx.py, surgery.py, compile.py
├── app1-single-stream/    session 3: main.cpp, CMakeLists.txt, app1.conf
└── app2-dual-stream/      session 4: main.cpp, CMakeLists.txt, app2.conf

Build output (build/), the model compilation working folder (model-compilation/work/) and model archives (*.tar.gz) are not tracked. Your own builds and compiles create them. Get the prebuilt archives from the instructor.

References

About

Content for Neat workshop

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages