Describing Images Using Semantic Modeling with Attributes and Tags

Explore image description techniques through semantic modeling using attributes and tags in this comprehensive lecture by Mahdi M. Kalayeh from the University of Central Florida. Delve into topics such as weighted multi-view non-negative matrix factorization, semantic segmentation-based gating and pooling, and the analysis of selfie popularity. Examine advanced concepts including Fisher vectors of Gaussian distributions and mixture models, as well as deep learning architectures like Inception-V3, DenseNet, and Deep Convolutional GANs. Gain insights into computational complexity analysis and future research directions in the field of image understanding and description.

Intro
Overview
Framework
Weighted Multi-view Non-negative Matrix Factorizatic
Summary
Methodology
SSG: Semantic Segmentation-based Gating
SSR: Semantic Segmentation-based Pooling
Experiments
Results
A unified view to SSP and SSG
Research Objectives
Selfie Dataset
Attribute Prediction
What Makes a Selfie Popular?
Sentiment-Popularity Correlation
Effect of Post-processing on Popularity
Introduction
Related Work
Proposed Method
Kernels from Generative Probability Models
Fisher Vector of Gaussian Distribution
Motivation
Fisher Vector of Gaussian Mixture Model
Mixture Normalization
Inception-V3
DenseNet
Deep Convolutional GAN
Computational Complexity Analysis
Conclusion
Future Work