Leveraging the Power of C++ for Efficient Machine Learning on Embedded Devices
Talkat CppCon 2023 · on Thursday, 2023-10-05· at 15:15· Cottonwood 2/3, 6700 North Gaylord Rockies Boulevard·
summary
C++ code uses TensorFlow Lite + OpenCV + GStreamer on a Raspberry Pi 4 Model B to run on-device ML: MobileNet (4MB) performs image classification (inference ~34 ms in a demo), a ResNet50 transfer-learning model (≈23MB) recognizes rock/paper/scissors with the small-trained model at ~65% overall and the larger dataset model near‑perfect on rock/paper but weaker on scissors. A live camera pipeline and a playable rock‑paper‑scissors game run on the CPU, inference latency scales with thread count up to core count, and misclassifications occur mainly from limited or mismatched training data.
AI-written summary · about
chapters (7)
- 0:00Introduction and motivation
- 13:32Image classification with C++
- 24:03Hand gesture recognition
- 31:44Live inference demo
- 35:51Rock Paper Scissors game
- 37:35Conclusion and future work
- 40:09Q&A session
related talks
39mLeveraging the power of C++ for efficient machine learning on embedded devices
27mMachine Learning on the edge using TensorFlow Lite
17mBuild a smart IoT device with TensorFlow Lite and Raspberry Pi
46mTensorFlow for Mobile Poets
42mOn-Device Machine Learning with Google AI Edge
32mAI for Mobile and IoT Devices: TensorFlow Lite
30mTensorFlow for Mobile Heroes
27mMachine Learning and Mobile