General information:
--------------------


The data avaiable are three Object class data sets: chairs, dogs, and humans. 
For each Object class data set there is one (or more) corresponding background data set, which makes the classification
task more challenging. 
For Chairs the corresponding is Furniture data. 
For Dogs, two background data sets are available: "easy animals" and "hard animals".  
For humans, the corresponding background is Sites data.

The data sets were and are used for object class recognition experiments. Specifically, they were used in
the following paper:

Efficient Learning of Relational Object Class Models, 
Aharon Bar Hillel, Tomer Hertz and Daphna Weinshall, 
ICCV 2005.

All the images are .jpeg images.



Specific data set information:
------------------------------

Chairs - contains 800 images of chairs of all kinds. The images were collected from the internet.
Most images contain a single chair, but some contain several chairs.
All the images contain chairs photographed from approximately the same view  (semi frontal, semi right frofile).

Furniture - Contains 500 images of 3 types of furniture:
		Images 1-200 are table images.
		Images 201-400 are bad images
		Images 401-500 are bookcase images.
	When meaningful (i.e. for most images), the images were flipped to the same view as the view of the chairs
	in the chairs data set.

Dogs - 	Contain 500 images of various Dogs. Dogs appear in approximately the same view (Semi frontal, semi right profile).
	However, view-point diversity is higher then in the Chairs data.
	The images were collected from the internet.

Easy animals - Contains 500 images of animals other then dogs. Specifically, the images contain
	51-100 Wild Cats
	101-150 Peacocks
	151-200 Reptiles
	201-250 Giraffes
	251-300 Monkeys
	301-350 Birds
	351-400 Rabbits
	401-450 Insects
	451-500 Parrots
	These animals do not have any fixed pose or view point.
	The images were collected from the internet.

Hard animals - Contains 500 images. Images 251-500 are identical to images 251-500 from 'easy animals' data set.
	Images 1-250 contain animals similar to dogs in an isomorphic pose to the pose and view point prominent in
	 the dogs images. Specifically, the images contain
	1-50 Horses
	51-100 Cows and Oxes
	101-150 Elephants
	151-200 Goats
	201-250 Dears
	The images were collected from the internet.

Humans - Contains 593 images of 25 persons. Images are 1600*1200 pixels, collected with a digital camera. 
	Each person was photographed in various poses and scales. The images are ordered according to persons 
	photographed. Each person was photographed at a diffeent background. 
	Specifically, for each person, images were taken as follows 
	
	Scale: Each person is photographed in at least 4 scales. The largest scale include the person from the knees up.
	Each subsequent scale is approximately 1.5 times smaller then the  previous. there are at least 2 images at each 
	scale.
	
	Viewpoint: The majority of the images are frontal. At least two (usually more) images are taken at half-profile
	view, both left and right.
	
	Pose: In 2-4 images subjects were instructed to wave their  hand in various poses.
	In 2-4 different images subjects were instructed to wave their legs, which usually leads to non-frontal pose.
	
	Partial occolusion: 2-3 images were  taken in varying partial occolusion conditions.

	Free exprerssion: In the last 4-5 images persons were instructed to choose any pose and condition they wish.
	This sometimes lead to very creative poses, ahich are rathjer hard to recognize.
	
Sites - Contains 593 images of the sites in which persons where photographed. 
	Images are 1600*1200 pixels, collected with a digital camera. 
	The number of photographes taken at each site is equal to the number of person photographes taken at that site.
	Most of the images describe locations, while some describe items found in the photography site.


For further information, if required, mailto aharonbh@cs.huji.ac.il