Overview

The tutorial slides can now be accessed here.

ECCV 2012 Tutorial T3 - Sunday 7 October, Florence, Italy (14:30 to 18:15, Room 1F Affari)

This course will introduce local feature detectors and descriptors as foundational tools in a variety of state-of-the-art computer vision applications. The first part of the tutorial will cover popular co-variant detectors (Harris, Laplacian, Hessian corners and blobs, scale and affine adaptation, MSER, SURF, FAST, etc.) and descriptors (SIFT, SURF, BRIEF, LIOP, etc.), with a particular emphasis on recent advances and additions to this set of tools. It will be shown how the various methods achieve different trade-offs in repeatability, speed, geometric accuracy, and applicability to different image contents in term of their performance in benchmarks and applications (tracking, reconstruction, retrieval, stitching, text detection in the wild, etc.). The second part of the tutorial will review software for computing local features and evaluating their performance automatically on benchmark data. In particular, two software resources will be introduced to the community for the first time: a novel extension to the popular open-source VLFeat library containing new reference implementations of co-variant feature detectors; and a novel benchmarking software superseding standard packages for the evaluation of co-variant feature detectors and descriptors.

Duration. 3 hours + hands on session.

Speakers. 
Andrea Vedaldi (U Oxford), Jiri Matas (Czech Technical University, Prague), Krystian Mikolajczyk (U Surrey), Tinne Tuytelaars (U Leuven), Cordelia Schmid (INRIA), Andrew Zisserman (U Oxford).

Hands on session. Karel Lenc (Czech Technical University, Prague), Michal Perdoch  (Czech Technical University, Prague), Daniele Perrone (University of Bern).

Sponsors.
PASCAL2 Logo
The novel computer vision software that will be presented during the workshop (VLBenchmarks and extensions of VLFeat) have been sponsored by the PASCAL Harvest 2012 programme.

Outline
  1. Introduction [2:30 2:45 (15m)]
    1. by Jiri Matas &  Andrew Zisserman 
    2. What feature detectors and descriptors are for: applications
  2. Feature detectors [2:45 3:30 (45m)]
    • by Jiri Matas and Krystian Mikolajczyk
    • What makes a good feature detector: evaluation axis
    • The classics
      • Corner measures: Harris Laplace/LoG/DoG/trHessian, Hessian (meaning detHessian), SMM (second moment matrix), Harris
      • Scale invariance: corner measure + LoG: Hessian-Laplace, Harris-Laplace
      • Affine adaptation: Hessian-Affine, Harris-Affine
    • Recent developments: SURF, FAST
    • Comparisons and discussion
  3. Feature descriptors [3:30 4:15 (45m)]
    • by Tinne Tuytelaars, Cordelia Schmid & Krystian Mikolajczyk
    • What makes a good feature detector: evaluation axis
    • The classics: SIFT
    • Recent developments: SURF, LIOP, BRIEF
    • Comparisons and discussion
  4. Break [4:15 4:30 (15min)]
  5. Feature implementations (software) [4:30 5:00 (30min)]
    • by Andrea Vedaldi
    • VLFeat: SIFT and covariant detectors
    • Other open source software: OpenCV
  6. Benchmarking: theory and software [5:00 5:30 (30min)]
    • by Andrea Vedaldi
    • Standard affine invariant testbed
      • Repeatability measure details
      • Matching measure details
    • Application testbed: image retrieval
  7. Hands on session training [5:30 6:15 (45m)]
    • by Karel Lenc, Michal Perdoch, and Daniele Perrone
    • Hands on experience: 
      1. Download feature extraction software VLFeat: vlfeat-0.9.16-bin.tar.gz
      2. Download benchmarking software VLBenchmarks: vlbenchmakrs-1.0-beta.tar.gz
        • Repeatability of VLFeat affine covariant features: