CPSC 533C Assignment 1:
Andrew Carbonetto
Both images come from the following paper on object recognition through a statistical learning
process. The learning process consists of a training set of images and "keywords". Objects are
deemed "recognized" when they can be consistently found correctly in a test set of images (without
"keywords").
image source:
Peter Carbonetto, Nando de Freitas and Kobus Barnard. A Statistical Model for General Contextual
Object Recognition. European Conference on Computer Vision, May 2004.
Good Visualization
Although this image is far from perfect, I consider it a good image because it is instantly
and easily understood. Without a full description, many computer scientists with a vague understanding
if machine learning and vision (such as myself) can understand that a computer simulation/learning
process is being performed upon a "training" and "test" set.
The image itself is well partitioned. The given image being rather colorful, and the outputted
image being block-colored. The image is also visually pleasing, since it isn't cluttered.
It is rather easy to see that the simulation still runs with many errors.
Bad Visualization
This Second image is from the same paper. The legend states that a darker grid square corresponds to
a stronger association.
At first glance, the grey scale colors are confusing. What's really awful is how compressed the data
is. The data is likewise ordered alphabetically, instead of being sorted into groups of related keywords
(such as snow and polar bear). It is also difficult to see if the data was successful or not based
entirely on this image.