Scene Attribute Dataset

The Scene Attribute dateset (SceneAtt) is used for studying shared attribute appearance models, and scene spatial configurations. The dataset (1225 images of 256*256 pixels in size) is selected from LabelMe Outdoor Scene dataset [1] and SUN Attribute dataset [2] which only contains the outdoor scenes such as "ocean", "open country" and "street". And to tolerate more objects (e.g., wild animals), we also add some images collected from Google images and Flickr. We ask people to give descriptions for each image. Then we obtain an attribute set which contains 17 noun attributes and 30 noun+adjective attribute pairs in total. The average number of noun-adjective pairs attached to each image is 3. And for each noun-adjective pair, the average image number is 96.

Attribute	Number of images	Attribute	Number of images	Attribute	Number of images
black-bison	51	dusthaze-sky	200	tree	286
elephant	59	rocky-mountain	96	green-field	385
white-horse	45	snowy-mountain	71	snowy-field	51
brown-horse	71	birdseye-building	15	yellow-filed	190
scarlet-ibis	54	perspective-building	117	road	162
black-ibis	29	front-building	46	tower	67
white-ibis	42	red-flower	29	blue-ocean	136
blue-sky	426	purple-flower	23	green-cliff	59
overcast-sky	179	pink-flower	17	black-cliff	43
cloudy-sky	170	sand	74	waterfall	58

Image	Text Description	Ground-truth		Image	Text Description	Ground-truth
	dust-haze sky; black bison; green field				overcast sky; birdseye building
	dust-haze sky; elephant; green field				blue sky; tree; yellow field
	blue sky; tree; road				tree; brown horse; green field
	blue sky; tree; white horse; green field				green field; scarlet ibis
	blue sky; black ibis				blue ocean; white ibis
	blue sky; rocky mountain; tree				blue sky; snowy mountain
	cloudy sky; tree; blue ocean				dust-haze sky ; purple flower
	dust-haze sky; front building; green field				road; perspective building
	tree; tower; green field				black cliff ; waterfall

Paper

dataset

S.Wang, J.Joo, Y.Wang and S.C.Zhu. Weakly Supervised Learning for Attribute Localizaiton in Outdoor Scenes. CVPR, 2013.

[1] C.Liu, J.Yuen, and A.Torralba. Nonparametric scene parsing: label transfer via dense scene alignment. CVPR, 2009.
[2] G.Patterson and J.Hays. Sun attribute database: discovering, annotating, and recognizing scene attributes. CVPR, 2012.