Vision Texture 1.0 (3/25/95) ============================================================================= This readme file contains a text version of the documentation maintained on the Vision Texture homepage. To obtain the latest information about VisTex check the homepage: http://www-white.media.mit.edu/vismod/imagery/VisionTexture/vistex.html FILES ------------------------------------------------------------------------------ This distribution of VisTex includes these files: Images - contains symbolic links to images in the FLAT directory Images/Reference/ - reference textures arranged by category Images/Scenes/ - texture scenes Images/VideoTextures/ - NOT AVAILABLE IN THIS RELEASE Images/Orbits/ - NOT AVAILABLE IN THIS RELEASE FLAT - image files arranged by size photobook - photobook annotation information DATABASE DESCRIPTION ----------------------------------------------------------------------------- Vision Texture is a collection of texture imagry for the computer vision community. Unlike other texture collections, the images in VisTex do not conform to rigid frontal plane perspectives and studio lighting conditions. Instead, the images that make up VisTex are taken from many sources including examples of from common ("real-world") photography and video. The database has 4 main components: - Reference Textures .. 100+ homogeneous textures in frontal and oblique perspectives. - Texture Scenes ...... Images with multiple textures. ("real-world") scenes. - Video Textures ...... Sequences of temporal textures. (NOT AVAILABLE IN ver 1.0) - Video Orbits ........ Images within a common projective group. (NOT AVAILABLE IN ver 1.0) OBTAINING THE DATABASE ------------------------------------------------------------------------------ Using FTP (NOTE: there is no longer an FTP site; use the Web directions below. -Lab Webmaster) The distribution site for the Vistion Texture database is: whitechapel.media.mit.edu Because of the large size of the database (240 megs uncompressed), the distribution is broken up into several smaller files for transfer via ftp. The directory pub/VisTex contains these files. Transfer all the compressed tar files along with the extraction script (buildVisTex) to your local machine. The script buildVisTex uncompresses the files and restores the VisTex directory structure. The compressed files for VisTex are large. We strongly suggest that the database be downloaded during none peak hours. Using the World Wide Web You can use your favorite net browser to download the database: http://vismod.media.mit.edu/pub/VisTex/ The compressed files for VisTex are large. We strongly suggest that the database be downloaded during none peak hours. Installing the Database To install VisTex you will need the tar and compress unix file utilities. The buildVisTex script unpacks the database files using uncompress and tar. The script should be run from the directory that contains the compressed tar files. ABOUT THE VISION TEXTURE DATABASE ----------------------------------------------------------------------------- About the Sections The images that make up the VisTex database are stored in a flat directory structure according to image size (VisTex/FLAT). The 4 sections of the VisTex database are accessed through a directory structure that contains symbolc links to the images in the flat directories. By using symbolic links, images may appear in several categories, but are stored once. The images in this sense are seperated from the classification system. Reference Textures In the Reference Texture section the images are grouped according to what they are. The (VisTex/Images/Reference) directory contains groups such as grass, water, brick, buildings, clouds, . . . etc. Within these directories are the symbolc links to the image files located in the FLAT directory. Texture Scenes The Texture Scenes section is arranged so that the links to all images relating to a given scene are in the same directory. The name of the scene dirctories come from the dominant textures in the scene. Within each scene directory (i.e. VisTex/Images/Scenes/GrassPlantsSky) there is a file / symbolic link named context1.ppm. This file is either a 786x512 or 512x786 image depending on the orientation of the scene. The other files in the directory are texture patches (usually homogeneous) taken from within the scene. While these detail patches are visible within the context scene, they are taken from alternate images so that the perspective and lighting varies slightly between the context and detail image. Video Textures This section is not yet available for release in version 1.0. Video Orbits This section is not yet available for release in version 1.0. GENERAL INFORMATION ------------------------------------------------------------------------------ File Formats and Image Size All images in the Vision Texure database are stored as raw ppm (P6) files. Currently there are 2 standard sizes for images in VisTex: 512x512, and 128x128. All images in the Reference Texture section and the texture patches from Texture Scenes are available in these sizes. The context images for the Texture Scenes section are 786x512 and 192x128. Vertically oriented images are available at 512x786 and 128x192. In most cases the textures in the database are available by special request up to 3072 x 2048. Annotations Annotations for the images are stored in the header of each ppm image. The current annotations for the database include: - content - lighting - perspective - Brodatz replacement number - source - notes The ppm header this image of tree bark looks like this: P6 # CONTENTS tree bark # LIGHTING daylight direct right # PERSPECTIVE frontal plane # BRODATZ D13 # SRC PCD 3174 2022 0103 #46 # NOTES - 512 512 255 Comment lines in ppm files begin with #. VisTex uses comment lines to store image annotations. The capatalized word after # is the annotation tag. Anything following the tag is considered to be the value of the tag. A list of current annotation tags and their possible values is available. For more information about the ppm file format see the browsers section in this document for information about the PBMPlus toolkit. BROWSERS AND SUPPORT PROGRAMS ---------------------------------------------------------------------------- Photobook The preferred browser for the Vision Texture database is Photobook. Photobook is an interactive browser of image databases which uses machine vision to perform queries based on content. Besides providing the capabilities to search on keyword annotations, photobook allows the user to perform queries based on "content", as defined by which vision algorithm is used. For example, images may be queried based on low-level features such as coarseness, color histogram, directionality, or periodicity. Suitable vision algorithms may even allow queries based on higher-level features such as the presence of certain objects or the spatial relationships of those objects. The VisTex distribution site maintains the latest additions to Photobook's precomputed image data for the database. This data allows the user to create custom groupings and indicies for the database. XV Image Viewer XV is John Bradley's popular image browser for X. The program allows for simple browsing as well as basic image manipulation (filtering, color adjustment, and scaling). To obtain a copy of XV see the XV distribution site (ftp.cis.upenn.edu), or download the tar and compressed distribution (xv-3.10a.tar.Z) of XV directly from this page. The compressed XV file is 3.4 megs. To load images into XV change into the directory with the images and run: xv *.ppm. Since annotations are stored in the ppm file headers they are available through the "Image Comments" command under the windows menu in XV. PBMPlus Image Toolkit PBMPlus is a popular set of image conversion and manipulation programs by Jef Poskanzer. The various programs that make up the toolkit work on the portable pixel map (ppm) file format which VisTex uses. While the kit does not provide a browser, it does have many useful routines for handling the images in VisTex. The PBMPlus Toolkit is available via anonymous ftp from ftp.ee.lbl.gov. The file pbmplus10dec91.tar.Z contains the source for the Toolkit which needs to be compiled. To compile PBMPlus to work with TIFF images Sam Leffler's TIFF Software package is required. COPYRIGHTS ----------------------------------------------------------------------------- Copyright 1995 by the Massachusetts Institute of Technology. All rights reserved. Developed by Rosalind Picard, Chris Graczyk, Steve Mann, Josh Wachman, Len Picard, and Lee Campbell at the Media Laboratory, MIT, Cambridge, Massachusetts. This distribution is approved by Nicholas Negroponte, Director of the Media Laboratory, MIT. Permission to use, copy, modify, publicly display and distribute this database and/or the individual images in electronic or print form is hereby granted without fee provided that (i) all use of the images is solely for educational and research purposes; (ii) all distribution of the images is at no-cost, except for recovery of cost of the media; and (iii) the copyright notice and the authors name(s) appear on all copies copies, and this permission notice is included with each distribution. You may also publish the images in scholarly journals to report research results without further permission from MIT. Any other commercial distribution or publication of the images, in original or modified form, requires the prior written permission of MIT. The name of MIT and/or the Media Laboratory may not be used in advertising or publicity without the prior written permission of MIT. ------------------------------------------------------------------------------ 3/25/95 - Chris Graczyk - cgraczyk@media.mit.edu