Watch a comprehensive thesis defense presentation on image understanding and context utilization in computer vision. Explore innovative approaches to scene identification, object detection, and image representation using Top-Down Visual Tree (TDVT) models. Learn about the incorporation of context through random walk techniques and their application in improving video analysis. Discover experimental results comparing traditional object detection methods with the proposed framework, and gain insights into future research directions in image representation and understanding.
Contextual Image Understanding and Scene Identification in Videos - Thesis Defense
University of Central Florida via YouTube
Overview
Syllabus
Intro
Image Understanding using context
Finding the Shared Context
Improving understanding on videos usin context • Scene Identification
Image Representation: Top-Down Visual Tree (TDVT)
How to learn TDVT representations?
A. Dataset Parsing (pre-processing)
Inference
A Labels: Top-down Tree LSTM
Model: Top-Down Tree LSTM
B Training 4 Classifiers: Is the node having an X edge? Train 4 classifiers
Evaluation: User study
Object detection Vs proposed approach
Evaluation: Object detection across datasets
Proposed Framework
Words weights for each query image
Incorporation of Context (Random Walk)
Incorporation of Context Random Walk
Experimental setup
Qualitative Results Image Retrieval Application from multiple
Quantitative results
Improving Scene identification on Videos
Improve scene Identification
Object detection on videos using context
Improve Object Detection
Experiments (scene identity)
Experiments (object detection)
Conclusions
Future Work (Image representation)
Taught by
UCF CRCV
