Analysis of Large-Scale Visual Recognition - Bay Area Vision Meeting

Overview

Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!

Grab it

Explore the evolution and challenges of large-scale visual recognition in this 40-minute Meta conference talk from the Bay Area Vision Meeting. Dive into the IMAGENET Large Scale Visual Recognition Challenge (ILSVRC) from 2010-2012, examining its diverse object classes and two main tasks: Classification and Classification + Localization. Analyze the performance of various systems, including SuperVision (SV) and VGG, across different scales and object types. Investigate the impact of clutter, texture, and scale on recognition accuracy. Conclude with insights into the ILSVRC 2013 large-scale object detection task, gaining a comprehensive understanding of the field's progress and future directions.

Syllabus

Intro
Large-scale recognition
IMAGENET Large Scale Visual Recognition Challenge (ILSVRC) 2010-2012
Variety of object classes in ILSVRC
ILSVRC Task 1: Classification
ILSVRC Task 2: Classification + Localization
ILSVRC (2012)
Chance Performance of Localization
Level of clutter
SuperVision (SV)
Difference in accuracy: SV versus VGG
Cumulative accuracy across scales Classification only
Textured objects (ILSVRC-500)
Textured objects (416 classes)
Localizing textured objects (416 classes, same average object scale at each level of texture)
ILSVRC 2013 with large-scale object detection