Overview
Learn about SILCO, a novel approach to object detection that requires only a few labeled images, in this 32-minute lecture from the University of Central Florida. Explore the concepts of few-shot classification, weakly supervised detection, and object co-detection. Discover how SILCO differs from traditional methods and delve into its key components, including the backbone architecture, spatial similarity module, and feature reweighting module. Examine the training process, experimental results, and comparative evaluations. Gain insights into the effectiveness of SILCO across various object sizes and scenarios, and understand its potential impact on computer vision applications.
Syllabus
Intro
Table of Contents
Problem Intro
Few Shot Classification
Weakly Supervised Detection
Object Co-Detection
How SILCO is Different
Overview
Backbone & Final Detection
Global Average Pooling: Baseline
Spatial Similarity Module
Method: Feature Reweighting Module Graph Convolutional Networks
Training
Results Experimental Setup
Ablation
Effect of Support Images
Effect of Object Size
Success and Failure Cases
Comparative Evaluation
Summary and Conclusion
Taught by
UCF CRCV