Deep Learning in Computer Vision
summary
Shows a working pipeline that starts with handcrafted 3x3 filters to detect edges, frames and glasses, then switches to learned filters by training a convolutional net in TensorFlow with NumPy and OpenCV helpers and reports about 80% accuracy for a tiny net and near 100% on a small validation set after scaling. Demonstrates transfer learning: strip a celebrity classifier to get face embeddings, train a tiny 3-layer comparator in Keras on thousands of pairs, and runs a live demo that is flaky but largely successful.
AI-written summary · about
chapters (11)
- 0:00Deep learning intro
- 1:10Science fiction meets modern AI reality
- 6:57Traditional to convolutional vision
- 11:35Glasses detection neural net
- 19:53ReLU and training power
- 24:11Handcrafted filters die
- 27:26Deep learning challenges
- 29:12One-shot learning
- 33:05Face recognition demo
- 43:32Mind-reading AI ethics
- 45:24Q&A
related talks
26mComputer Vision in the Era of Deep Learning
42mDeep Learning for Computer Vision
04mUsing Computer Vision to Understand Big Visual Data
01h 29mComputer Vision: An (Un?)Expected Journey, with Keras and Tensorflow
35mDeconstructing Deep Learning
59mDeep Neural Networks for Video Applications
54mIntroduction to Understanding Deep Learning
55mA Tour of Deep Learning With C++