VideoCapsuleNet - A Simplified Network for Action Detection
Offered By: University of Central Florida via YouTube
Course Description
Overview
Explore a simplified network for action detection in this 22-minute lecture on VideoCapsuleNet. Delve into the fundamentals of Capsule Networks, computer graphics, and inverse graphics before examining the architecture of VideoCapsuleNet. Learn about convolutional capsule layers, capsule pooling, and the encoder structure. Analyze the training process, action localization accuracy, and qualitative results for entire videos. Investigate the effects of capsule masking and various ablation studies, including coordinate addition, skip connections, convolutional layers, losses, and reconstruction. Conclude with insights from synthetic dataset experiments to gain a comprehensive understanding of this innovative approach to video action detection.
Syllabus
Intro
Overview of Capsule Networks
Computer Graphics
Inverse Graphics
Capsules
Conventional Convolutional Layers
Convolutional Capsule Layers
Two Simplification
Capsule Pooling
Current Video Action Detection Network
VideoCapsuleNet Architecture
Encoder
VideoCapsuleNet Training
Action Localization Accuracy
Qualitative Results - Entire Videos
Effects of Capsule Masking
Ablations: Coordinate Addition
Ablations: Extra Skip Connections
Ablations: # of Convolutional Layers
Ablations: Losses and Reconstruction
Synthetic Dataset Experiments
Taught by
UCF CRCV
Tags
Related Courses
الشبكات العصبية والتعلم العميقDeepLearning.AI via Coursera Machine Learning: Create a Neural Network that Predicts whether an Image is a Car or Airplane.
Coursera Project Network via Coursera Neural Network Programming - Deep Learning with PyTorch
YouTube Computer Vision with GluonCV (Traditional Chinese)
Amazon Web Services via AWS Skill Builder Neuronales Netz von Scratch
Coursera Project Network via Coursera