Go to content Go to navigation Go to search

ICCV - Day 2

Workshops start on the second day. There were 8 workshops running concurrently, and I was particularly interested in the PASCAL Visual Object Classes Challenges 2011, and really had to go check out what the CV community had been up to in the area of object classification.

There were three main competitions, one each for

  • image classification — does the image contain an instance of this class?
  • object detection or localization — find the bounding box for each instance of the class
  • object segmentation of class region of interest

There were also three ‘taster’ competitions, which were not as popularly participated as they were very challenging,

  • determination of person body parts layout — where are this person’s head, hands and feet?
  • still-image action classification — is this person running/playing an instrument, etc.?
  • large scale recognition using ImageNet (in which a subset of >1 million images of >1,000 categories were utilised)

Instead of trying to recall what were the methods that performed reasonably well in the challenge, I thought it might be more useful if I had an overall idea of the basket of methods that the majority of participants seemed to have tried to some good measure of success.

Here’s a list of stuff that 1) I do not entirely know, or know very little about, and 2) that keeps re-appearing in many of the methods employed for object classification (and possibly re-appear many-fold more in the next few days):

Bag-of-words (visual words/vocabularies), spatial pyramids, sparse coding, locality-constrained linear coding (LLC), latent structural SVM, multiple kernel learning (MKL).

I guess I have not been getting up-to-date for quite a number of years now, many thanks to the lack of opportunities to be in a really top conference, and obviously I have not been actively scouting and nitpicking other areas of CV outside my PhD work.

Had one of the best (or with a high likelihood of being the best) paella marinera with Josiah at a neat and cozy bistro, which had an unassuming name of Rias de Vigo. Separate post on the food later.

Leave your comments

  You can enclose *strong* for strong or _emphasis_ for emphasis
Find out more at Textile Help
This is not SPAM

Live comment preview


to be posted on 20 September 2017, 17:25

Subscribe to Tulips by Email

 » This article » 

Back to Archive list