Using Computer Vision to Solve Interesting Problems
summary
Builds custom detectors with YOLOv7 using Frigate as the local video pipeline, Anaconda and LabelImg for setup and annotation, training on roughly 100–200 labeled images and about 1.3 hours on a laptop GPU for an initial run. Toy snakes dragged on strings produce usable training images but real €200 security cameras are too low resolution so snake detection fails in practice. Shark detection trained on ~200 drone screenshots works much better, demo detects sharks in water but is not yet production‑grade and needs more data and better cameras.
AI-written summary · about
chapters (10)
- 0:00Introduction to fun CV
- 1:05Snake detection idea
- 3:39Training AI with toy snakes
- 6:52Testing with real security cam
- 9:18Next steps and challenges
- 10:24Shark detection inspiration
- 11:52AI drone patrols concept
- 13:43Live shark detection demo
- 15:10Future product potential
- 15:30Q&A session
related talks
23mBirding with Python and Machine Learning
32mApplying Deep Learning for Object Detection from a Telecom Perspective
45mWhat if AI were the solution to understanding sign language?
01h 15mTrain Object Detection with Small Datasets
24mAI in maritime - Autonomous ships
21mLeveraging Python for Real-Time Image processing
23mBuilding high-quality computer vision models using only a few examples
57mDrones & AI - What's all the buzz about ?