|Description: This is a new data set based on the CalTech 101 image annotations.
Each image in the CalTech 101 data set includes a high-quality polygon outline
of the primary object in the scene. To create the CalTech 101 Silhouettes data
set, we center and scale each outline and render it on a DxD pixel image-plane.
The outline is rendered as a filled, black polygon on a white background.
Many object classes exhibit
silhouettes that have distinctive class-specific features. A relatively small
number of classes like soccer ball, pizza, stop sign, and yin-yang are
indistinguishable based on shape, but have been left-in in this this version
of the data.
Post-Processing: Outlines that result in images that are either more than 90% white or black are dropped from the data set. The train/validation/test split is a stratified sample since the classes are badly imbalanced. Given Nc instances from class c, we put min(3/5*Nc,100) instances from that class into the training set, so each of the 101 classes has at most 100 training instances. The minimum number of training instances per class is around 20. The remaining instances are split evenly between validation and testing. The validation and test sets for each class have between 6 and 400 instances. It makes sense to use class-balanced prediction accuracy, as for the standard CalTech data set, since the test and validation sets are badly imbalanced and some classes may be much easier to predict than others.
Data Set Images: The images below shows the first 10 instances of each class. We show the outlines rendered as 28x28 images as well as 16x16 images. Even at 16x16 pixels, many classes still exhibit distinctive class-specific features.
Data Set Matlab Files: