Sample Images for CS
664 - Computer Vision
Prof. Dan Huttenlocher
Fall 2003
There are 8 pairs of images, each named xxxx_1.ppm and xxxx_2.ppm. Please run your implementation
of part #1 on all 8 pairs, and include the output (motion vector estimate) for each pair in your report.
(Hint: The ideal motion vector between boxes_1.ppm and boxes_2.ppm is a 0.5 pixel shift to the left.
The ideal motion vector between moreboxes_1.ppm and moreboxes_2.ppm is a 0.5 pixel upward shift.)
There are 4 pairs of images, again named xxxx_1.ppm and xxxx_2.ppm.
10/23/03 Note: Don't worry if your algorithm doesn't work well on cayuga_1.ppm and cayuga_2.ppm.
I included them in the zip file by mistake. I've included an additional pair (mcfaddin_1 and mcfaddin_2)
as a substitute.
Important note: Contrary to the information on the assignment sheet, some of these
images require using a Gaussian Pyramid with more than 5 levels. You may
want to make the number of levels in the Gaussian Pyramid a parameter of
your algorithm. I (David) found that 10 levels worked well for all images.
(Hint: Start with start_1.ppm and start_2.ppm, because they have the least
amount of motion and no distortion.)
All images are in .PPM format. The
libraries recommended on the course
web site support reading and writing .ppm files. The .ppm format is
simple enough that you can also write your own I/O routines instead, if you
wish. The cost of this simplicity is that .ppm files do not support compression and hence
can be very large.
Note that a few of the test images do not contain stop signs. Running your
algorithm on these images may help you identify potential false
alarms (e.g. other objects that are incorrectly identified as
stop signs by your program) and revise your object model accordingly.
Images for Assignment 2
These are the test images for Assignment 2, the image motion tracker.
Images for Assignment 1
Below you'll find images to use as test data for assignment 1, the stop sign
detector.
We recommend that you start with image
set #1, which contains relatively easy images (i.e. the signs are
prominent, parallel to the image plane of the camera, roughly uniform
in size and orientation, etc.). We'll add another image
set later that will be a little more challenging. Feel free to use
your own images for testing as well.
Note: The above archive of ppm files is very large. Alternatively, you
can download a zip archive
of the images in .jpg format (~2 MB), but you will have to
convert the .jpg files to .ppm files on your local machine.
These are sample output images generated by running a subset of image set #1 through David's (TA's)
implementation of assignment #1. Of course, these results are just samples; your detector will undoubtedly produce different
outputs
due to different model parameters, etc.
Image set #2 contains a few more difficult images to test the robustness of your model.
Note that some of these images are synthetic, but hopefully they will still be useful
for evaluating the strengths and weaknesses of your detector.