Over the past few months, we’ve helped many developers kick-off Hailo app development with DeGirum PySDK. A common request was a clear learning roadmap—how to explore features in the right order and build up capability, step by step.
We’ve turned that feedback into a progressive set of guides that go from simple to advanced and cover practical, real-world workflows.
What’s inside (roadmap at a glance)
Basics (start here)
-
Before you begin — prerequisites (venv, HailoRT/driver check, device discovery) and an optional workspace token step to streamline model access.
-
First inference — a 1-minute quick win to confirm everything works end-to-end.
-
Specifying a model — how
ModelSpecdefines the model, where artifacts live, and where inference runs. -
Running inference — the three core paths: single images, video/streams (via a convenience wrapper), and folders.
-
Inference results — the
InferenceResultsobject: inspect structured outputs, visualize overlays, save, or stream. -
Measuring performance — baseline checklist, per-frame timing, throughput, and regression tracking.
Intermediate guides
-
Class filtering — focus outputs on the labels you care about.
-
Object tracking — attach a tracker to assign persistent IDs and enable motion analytics.
-
Zone-based counting — polygon zones for traffic/retail analytics, with interactive zone editing.
-
Tiling— four tiling strategies to boost small-object recall and merge results cleanly.
-
Model properties — device/runtime selection, input preprocessing, output postprocessing, and overlay controls.
Advanced
-
Custom video sources — plug in OpenCV, GStreamer, PiCamera2, SDKs, or any generator with
predict_batch. -
AI Server — use the DeGirum AI Server with cloud or local models; pointers for monitoring via
HAILO_MONITOR.
Start here
DeGirum Docs for Hailo — quickstart + the full guided journey.
If there’s a use case you’d like us to add (e.g., multi-stream, multi-models, smart parking, industrial inspection, healthcare), reply below—we’re expanding the series based on community feedback.