Project 4: Class Recognition Using
Discriminative Local Features
Description
•
Implement
the part based method for object recognition, described in “Class Recognition Using Discriminative Local
Features”, by G. Dorkó, C.
Schmid. Implement only likelihood ranking.
•
Test
it on
•
Compare
the performance of the algorithm using different point detectors: Saliency
Scale, Saliency Affine, Harris-Laplace scale, Harris-Laplace Affine (The code
for point detectors is provided).
•
Compare
the performance of the algorithm with original SIFT and with SIFT without
rotation invariance. The initial code for SIFT is provided, but should be
edited to remove rotation invariance.
Results
The classifier trained on
specific category, say airplanes, should be tested on airplanes images and on
images of general background. The results should be organized in Receiver
Operating Characteristic (ROC). This curve show correct detections as a function
of incorrect detections. Then calculate an Equal Error Rate (EER) as
p(True Positive) = 1 -
p(False Positive),
True positives are images of
airplanes that were classified as containing airplanes. False positives are
images of background that were classified as containing airplanes.
Do the same for all the
categories.
Compare EER for different
choices of interest point detector. Note that the algorithm has a parameter p
that should be adjusted when changing the interest point detector.
Data set
Oxford dataset:
containing faces, airplanes, cars, motorbikes ( The background collection is
not very good there)
Background
images:
In GRAZ_01 download
no_bike_no_person.zip
In GRAZ_02 download
none.zip
Interest Point Detectors
SIFT Software
·
Matlab
Useful Links
Class lecture on object
detection
"Distinctive image features from
scale-invariant keypoints" by David Lowe
“A
comparison of affine region detectors”, K. Mikolajczyk, T. Tuytelaars, C.
Schmid, A. Zisserman, J. Matas, F. Schaffalitzky, T. Kadir and L. Van Gool.
“Scale and affine invariant interest point
detectors” K. Mikolajczyk, C. Schmid, IJCV, Volume 60, Number 1 - 2004
Status
Claimed by Aviv and Pavel