Birding with Python and Machine Learning
summary
Bootstraps bird detection with transfer learning: uses Tiny YOLO (VOC-trained) via Darkflow to detect birds, then labels species with labelImg and augments ~200+ images before retraining. Hardware ranges from USB webcam + Raspberry Pi streaming to a Jetson TX1/TX2 with SSD for on-device inference, housed in a wooden enclosure with 3D‑printed camera mounts. Recording logic uses a pre-buffer and writes 10s before and after detection and cron + ffmpeg to batch MP4s for later labeling.
AI-written summary · about
chapters (15)
- 0:00BIRDING WITH PYTHON AND MACHINE LEARNING
- 1:14NEURAL NETWORKS ARE SOFTWARE 2.0
- 2:50MODERN MACHINE LEARNING IS BIG DATA AND BIG COMPUTATION
- 4:44HACKERS AND HOBBYISTS NEED NOT APPLY TO MACHINE LEARNING?
- 9:27ROBOTIC PLATFORM TO EXPLORE CROW INTELLIGENCE
- 9:43PROBLEM OF DETECTING BIRDS TRANSFERABLE TO OTHER SPECIES
- 13:12USE TINY YOLO MODEL TRAINED ON VOC PORTED TO TENSORFLOW
- 13:43BOOTSTRAP BY FIRST DETECTING BIRDS
- 14:49DEPLOY AND COLLECT DATA
- 17:16BOOTSTRAPPING MORE TRAINING DATA
- 18:31LABEL AND AUGMENT NEW DATA
- 19:24ADD NEW CLASSES FOR EACH NEW SPECIES YOU DETECT
- 19:57RETRAIN AND REDEPLOY
- 21:36DEPLOY IN WILD?
- 22:33EXAMPLE WITH CLOTHING, BRANDS, EMOTION, AGE DETECTION
related talks
28mChasing birds with code
10mDeep Learning in the Wild: Embedded PyTorch for...
21mUsing Computer Vision to Solve Interesting Problems
35mAI is for the birds
59mOpening Keynote
40mWho's singing? Automatic bird sound recognition with machine learning
21mLeveraging Python for Real-Time Image processing
32mApplying Deep Learning for Object Detection from a Telecom Perspective