Labeled databases for object detection

List compiled by Kevin Murphy
with contributions from David Lowe, Peter Carbonetto, Robert Sim.
Last updated 2 September 2005.

LabelMe: a database and web-based tool for image annotation
Bryan Russell, Antonio Torralba, Kevin Murphy and William Freeman
Submitted to Intl. J. Computer Vision (special issue on vision and learning), August 2005.

Generic classes

Name Objects Background Annotation #ex per class
LabelMe ~500 categories (indoor and outdoor objects) ~3000 labeled natural images Bounding polygons 50-500
MIT-CSAIL 107 categories (indoor and outdoor objects) Natural images Bounding polygons ~200
UIUC carSide Side views of cars Natural images Bounding boxes ~500
ETHZ Motorbikes, cars, cows Natural images Bboxes + seg masks ~100
Graz 02 Bicycles, cars, cows, pedestirans Natural images Segmentation masks ~300
Caltech Planes, motorbikes, faces, cars. Close up photos Bounding boxes 500-1000
Caltech 101 101 categories Close up photos None ~50-200
Pascal Cows, bicycles, people, etc. (meta site) Natural images Various Various
Wiry Objects (CMU) Chairs, bicycles, etc. Natural images Labeled edges ~100
UW 100s Natural images Captions ?
UBC 10s Corel segmented with Ncuts Captions ?
MIT-CBCL Frontal faces, cars, pedestrians None Isolated patches ~800
NORB (NYU) Toy cars, animals, etc (many angles) Synthetic cluttered backgrounds Isolated patches ~10
CMU/MIT frontal faces Frontal faces None Isolated patches ~1000
Frischholz Frontal faces (meta site) None Isolated patches ?

Specific instances of objects

Name Objects Background Annotation #ex per class
COIL 100 (Columbia) Highly colored man-made objects None Isolated patches 1
SOIL-47 (Surrey) 24 cereal-box-like objects in color None Isolated patches 1
ETH-80 8 classes, 10 instances per class, 41 views per instance None Isolated patches 10*41
UIUC Ponce Various (meta site) - - -