The Full Images for Natural Knowledge Caltech Office DB

We describe the image collection and labeling procedures of an office scene database. The office database is aimed at providing training and benchmarking tools for multiclass object detection in natural scenes. The described database is unique in the fact that high quality office images were collected following a protocol that (in theory) does not introduce a human bias in camera positioning and environment manipulation. 225 360deg office images all labeled with 14 objects are available. The objects are: LightSwitch, window, doorFrame, bookshelf, Monitor, whiteboard, Keyboard, chairBack, Television, roundClock, Laptop, cabinet, PictureFrame and openShelf. The use of the office database is conditioned by a contribution to the labeling effort.