Deep Neural Networks for Video Applications
summary
Shows concrete pipelines and results: uses pretrained convolutional nets (ImageNet->ResNet), transfer learning in Keras/TensorFlow to get 96.3% on ~3.5k clothing images; trains object detectors on COCO then fine-tunes for custom tasks. Builds license-plate system (detect plate → crop → char OCR) and a truck-bypass detector that sends real-time SMS alerts; implements person-tracking to count taxi passengers (example flagged 42 children). Adding temporal RNNs raises seal-detection from ~71% to ~93%.
AI-written summary · about
chapters (18)
- 0:00Start
- 12:24WHAT IS A NEURON?
- 13:49RANDOMLY INITIALIZED WEIGHTS
- 14:17(DEEP) NEURAL NETWORKS
- 14:29HOW DOES A NEURAL NETWORK LEARN?
- 14:47GRADIENT DESCENT
- 22:21IMAGE CLASSIFICATION
- 26:20Convolutional Nets Learn Hierarchical Features
- 29:05IMAGENET TOP-5 ERROR RATE
- 30:03Example: Use CNN to Classify Product Images
- 31:18USING A CNN AS A FEATURE EXTRACTOR
- 31:52Extracting Features from an Image
- 32:05Adding a New Classifier
- 36:02CUSTOM OBJECT DETECTION
- 36:30How to train your own Object Detector with TensorFlow's Object Detector API
- 47:00FACE SWAP
- 48:43POSE ESTIMATION
- 49:30DISTRACTED DRIVING DETECTION
related talks
42mDeep Learning for Computer Vision
28mObject Detection using Transformers and CNNs
08mDeep Transfer Learning for Computer Vision
38mDeep Learning for detection on a phone
21mRecognizing Human features using Deep Networks
01h 29mComputer Vision: An (Un?)Expected Journey, with Keras and Tensorflow
47mDeep Learning in Computer Vision
54mIntroduction to Understanding Deep Learning