Analysis of Large-Scale Visual Recognition

Olga Russakovsky (Stanford University)

COMPUTER VISION AND ROBOTICS SERIES

DATE: 2013-11-29
TIME: 14:45:00 - 15:30:00
LOCATION: RSISE Seminar Room, ground floor, building 115, cnr. North and Daley Roads, ANU
CONTACT: JavaScript must be enabled to display this email address.

ABSTRACT:
The growth of object detection datasets and the multiple directions of object detection research provide both an unprecedented need and a great opportunity for a thorough evaluation of the current state of the field of categorical object detection. Our paper titled "Detecting avocados to zucchinis: what have we done and where are we going?" (to appear in ICCV2013) strives to answer two key questions. First, where are we currently as a field: what have we done right, what still needs to be improved? Second, where should we be going in designing the next generation of object detectors? In this talk I will highlight the key findings.

To answer these questions, we performed a large-scale study on the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2012 data. First, we quantitatively demonstrated that this dataset provides many of the same detection challenges as the long-standing PASCAL VOC benchmark. Due to its scale of 1000 object categories, ILSVRC provides an excellent testbed for understanding the performance of detectors as a function of several key properties of the object classes. We conducted a series of analyses looking at how different detection methods perform on a number of image-level and object-class-level properties such as texture, color, deformation, and clutter.

For the most up-to-date methods, results and analysis of large-scale visual recognition we also encourage you to attend our "ImageNet Large Scale Visual Recognition Challenge 2013" workshop coming up on December 7th at ICCV2013 in Sydney: http://image-net.org/challenges/LSVRC/2013/iccv2013