Action Recognition and Video Classification Techniques - Lecture 19

Explore advanced concepts in computer vision and deep learning for action recognition in this comprehensive lecture from the University of Central Florida's CAP5415 course. Delve into topics such as sequences of images, action recognition using the UCF101 dataset, action detection with VIRAT, and video segmentation techniques. Learn about 3D convolution and its practical aspects in video processing. Examine various video classification variants and gain insights into recurrent neural networks, including vanilla RNN and Long Short-Term Memory (LSTM) architectures. Refresh your understanding of activation functions and discover how these concepts are applied to video activity recognition tasks.

Intro
Outline
Sequences of Images
Action recognition - UCF101
Action detection - VIRAT
Video segmentation
3D convolution
Practical aspects
Video classification variants
Recurrent neural network
Activation functions - recap
Vanilla RNN
Long-short term memory (LSTM)
Video activity recognition