NNStreamer: Efficient and Agile On-Device AI Application Development with Neural-Net Pipelines
summary
On-device pipeline framework that extends GStreamer by adding a tensor type and plugins for TensorFlow, TFLite, custom binaries and hardware backends, NNStreamer lets developers wire preprocessing, multiple models and renderers as one graph. A mobile demo bundles four models and a working pipeline that the uploader says can be written in five lines. Benchmarks show memory cut ~50% and throughput rising 20 -> 30 FPS versus a Python control, automatic synchronization and rate control save CPU and battery; a model-switch bug is noted and being tracked.
AI-written summary · about
chapters (10)
- 0:00Intro
- 0:23Al based Services
- 6:58Working Application
- 12:19Pipeline Code
- 15:24Multi-Platform
- 17:34API for Developers
- 19:53Extensibility
- 22:16Efficiency
- 25:31Ease of use
- 31:34Rate Your Tech Sessions
related talks
18mDeploying ML solutions with low latency in Python
46mFire Up Your Smartphone Neurons with On-Device Deep Learning
36mApplied AI on the edge
40mOn-Device Machine Learning for Android with MediaPipe
27mEmbedded Deep Learning Super Resolution on GStreamer using ONNX Inference Runtime
14mFine-tuning giant neural networks on commodity hardware with automatic pipeline...
38mNEURAL NETWORKS ON YOUR MOBILE PHONE. DIVING INTO THE NN API
32mAdvanced on-device ML made easy