Adventures in Puppy Training with A.I. and a Raspberry Pi
summary
Uses a camera plus Raspberry Pi and a Bluetooth speaker to give immediate praise or scold sounds; labels 274 videos (cleaned to 239) and expands them into ~298,000 temporal samples. Pose points come from DeepLabCut; feature selection focuses on tailbone and throat. A 237 MB LSTM gesture classifier runs alongside a 94 MB dog-pose model and 7 MB pad detector, reaches 97.8% accuracy, and moves from ~2 fps in Python to ~10 fps after rewriting in C++.
AI-written summary · about
chapters (13)
- 0:00Introduction and speaker intro
- 1:42Meet Tilly the puppy
- 5:40Problem and feedback challenges
- 7:48Machine learning basics
- 9:11Gathering training data
- 15:36Modeling approach with LSTM
- 20:37Pad detection model
- 22:45Raspberry Pi deployment struggles
- 25:54Optimizing latency with C++
- 26:37Adding feedback with sound
- 28:24Takeaways and lessons learned
- 31:54DIY AI book promotion
- 33:16Audience Q&A session
related talks
37mUnstructured Data Processing with a Raspberry Pi AI Kit and Python
40mPuppies or Muffins
36m"Lights, camera, AI!" - Automated sports videography
21mAI, Embedded System and Robotics
34mGCP for Data Scientists
43mBuilding Qrio, a Bot That Plays Videos for My Toddler
53mFROM ZERO TO DEEP LEARNING WITH SCALA
34mVoice-Controlled Lights on a Raspberry Pi