🚀 go-pugleaf

A number of images, volumes and sequences can be obtained by anonymous
ftp from isy-gw.liu.se (130.236.1.3). The data is located on the
directory /site/src/public/Packages/images/ where also a README-file
describing the data can be found.

I have appended the README at the end of this (semi-regular) posting.

	/Hakan

-------------------------------------------------------------------------------
H}kan B}rman	                   | Email: hobo@isy.liu.se
Linkoping University		   |       
Dept. of Electrical Engineering    |
Computer Vision Laboratory         |
S-581 83 Linkoping                 | Phone: +13 28 18 85
Sweden			           | Fax:   +13 13 85 26
-------------------------------------------------------------------------------

This is /site/src/public/Packages/images/README.
This file contains the following sections:
	How to convert .imf-files
	If you have problems
	Contents
	Description of the directorys
	Description of the images

This directory contains image-files (with the extension .imf) and
directorys (with the extension .ic).

All files with the extension .imf are binary. All other files are ascii.

How to convert .imf-files.
==========================
The image files are stored using the file format used at the Computer
Vision Laboratory at Linkoping University. The file format consists of
a header of 512 byte (describing the size, pixel type etc) followed by
the raw image data, pixel by pixel. You will need to do the following
two steps:

	1/ Get rid of the header info, eg by standard Unix tools such
	as dd (eg dd if=girl.imf of=girl.raw bs=512 skip=1)

	2/ Convert the raw image data to your personal image file
	format. The dimension of the images are given at the end of
	this text file.

If you have problems:
=====================
		List of trouble-shooters:
	hobo@isy.liu.se
	jowi@isy.liu.se
	mwe@isy.liu.se

CONTENTS:
=====================
	Text-files:
README		This file.
ChapelHill1.txt	How to get some more interesting volumes.
Stereo.txt	Calibration data for the stereo pair in calib.ic

	Image Catalogues:

barn.ic		colour image
breakfast.ic	stereo
calib.ic	stereo
frau.ic		sequence
heart1.ic	sequence
heart2.ic	sequence
knee.ic		volume

	Images:



=========================================
##### Description of the directorys #####
=========================================

barn.ic
=====================================================

Origin: Martin Hedlund
	ContextVision
	Struers Vision AB, Engelbrektsgatan 22
	Linkoping, SWEDEN

History: A colour image with the red, green and blue components stored
	in separate files. An outdoor scene from Linkoping, although
	it looks more like the countryside.

Format: A directory containing 3 images, (red,  green and blue) each file
	with 512 rows, 512 columns, 1 unsigned byte/pixel. 


Acknowledge: See origin.

breakfast.ic
=====================================================

Origin: CVAP (Computational Vision and Active Perception Laboratory)
	at KTH, Stockholm, Sweden. 

History: One stereo pair made with CVAPs Head-Eye Robot	System.
	(Pahlavan & Eklundh, Proc. SPIE Machine Vision and Robotics
	Conf., Apr 1992). [This scene has been refered to as `a
	typical Swedish breakfast table', but this is not true: Most
	of us do NOT have an icosahedron for breakfast. /hb]

Format: A directory containing 2 images, (left and right) each file
	with 512 rows, 384 columns, 1 unsigned byte/pixel. 

Acknowledge: Jan-Olof Eklundhs CVAP group at KTH, Stockholm, Sweden. 

calib.ic
======================================================
Origin:
 Wilfried Enkelmann	email: enk@iitb.fhg.de
 Fraunhofer Institut fuer Informations- und Datenverarbeitung IITB
 Fraunhoferstr. 1
 D-7500 Karlsruhe 1
 GERMANY

History: A stereo pair with ground truth for some of the pixels (see
	the text file stereo.txt). Note that the same camera setup is
	used in frau.ic.

Format: A directory containing 2 images. Each file
	with 512 rows, 512 columns, 1 unsigned byte/pixel. 

Acknowledge: See origin.

dodecROT_DC.ic dodecROTdB10 dodecROT_id
================================================
Origin: Computer Vision Laboratory, Link\"oping University, 
	Link\"oping, Sweden.

History: Synthetic 3D data designed for testing of 3D
	line/edge/orientation detectors, see:  
	@PhdThesis{h92,
	  author = 	{L. Haglund},
	  title = 	{Adaptive {M}ultidimensional {F}iltering},
	  school = 	{Link\"oping University, Sweden},
	  year = 	{1992},
	  address = 	{S--581 83 Link\"oping, Sweden},
	  month = 	{October},
	}

	10 lines distributed as the vertices of a dodecahedron. The
	"dB10" has white gaussian noise added to a SNR of 10 dB.
	SNR = 20 log(amp(signal)/sdev(noise)).

	The _id sequence contains the true optical flow if the volume
	is interpreted as an image sequence.


Format: Directories containing 64 slices (images), each file with 64
	rows, 64 columns, 1 unsigned byte/pixel except for the _id
	sequence which has 4 bytes/pixel. The first two bytes (a C
	short) represent the length of the flow vector. The last two
	bytes (a C short) represent the angle of the flow vector and
	should be mapped from [-32768, 32767] to [-PI, PI}.

Acknowledge: Leif Haglund.

fleet_div.ic fleet_div_id.ic fleet_trans.ic fleet_trans_id.ic
=============================================================
Origin: David Fleet email: fleet@qucis.queensu.ca 

History: These sequences are suitable for comparison of optical flow
	algorithms. (The `lenna' of optical flow?) They are described
	in: 
	@Book{fleet92,
	  author = 	"David J. Fleet",
	  title = 	"Measurement of image velocity",
	  publisher = 	"Kluwe Academic Publishers",
	  year = 	"1992",
	  note = 	"ISBN 0--7923--9198--5"
	}
	@InProceedings{barron92,
	  author = 	"J.L. Barron and D. J. Fleet and S. S.
			Beauchemin and T. A. Burkitt", 
	  title = 	"Performance of Optical Flow Techniques",
	  booktitle = 	"Proc. of the CVPR",
	  year = 	"1992",
	  pages = 	"236--242",
	  organization = 	"IEEE",
	  address = 	"Champaign, Illinois, USA",
	  OPTnote = 	""
	}

Format: Four different directories. _trans and _div contains the
	original sequences, while _trans_id and _div_id should contain
	the true optical flow (*). The trans sequence has 40 frames
	and the div sequence has 39 frames. All of the images have 150
	rows and 150 columns. The original sequences have 1 unsigned
	byte/pixel. The true OF sequences have 4 bytes/pixel. The
	first two bytes (a C short) represent the length of the flow
	vector. The last two bytes (a C short) represent the angle of
	the flow vector and should be mapped from [-32768, 32767] to
	[-PI, PI}.

Acknowledge: See origin.

(*):	The true optical flow data was generated here at the Computer
	Vision Laboratory but should be identical to the true flow
	used in \cite{fleet92,barron92}.



frau.ic
======================================================

Origin:
 Wilfried Enkelmann	email: enk@iitb.fhg.de
 Fraunhofer Institut fuer Informations- und Datenverarbeitung IITB
 Fraunhoferstr. 1
 D-7500 Karlsruhe 1
 GERMANY

History: One of many stereo sequences available from Karlsruhe. A
	woman(?) crossing the street.

Format: A directory containing 30 images, ie 15 stereo frames. Each file
	with 512 rows, 512 columns, 1 unsigned byte/pixel. 

Acknowledge: See origin.



heart1.ic
==================================================
Origin:  Tomas Gustavsson, Dept. of Applied Electronics, Chalmers
	 University of Technology, S - 412 96 Gothenburg, Sweden.

History: Cross-sectional ultrasound sequence revealing the chambers of
	the heart.  Note that the sequence is cyclic!

Format: A directory containing 16 images, each file with 128 rows, 128
	columns, 1 unsigned byte/pixel.

Acknowledge: Tomas Gustavsson, Dept. of Applied Electronics, Chalmers
	     University of Technology, S - 412 96 Gothenburg, Sweden.

heart2.ic
==================================================
Origin: See heart1.ic

History: See heart1.ic

Format: A directory containing 16 images, each file with 128 rows, 128
	columns, 1 unsigned byte/pixel.

Acknowledge: See heart1.ic

knee.ic
================================================
Origin: Data taken on the Siemens Magnetom and provided courtesy of
	Siemens Medical Systems, Inc., Iselin, NJ.  

History: A 127-slice data set of a human knee.  Complete slices are
	stored consecutively as a 256 x 256 array.  Pixels consist of 
	2 consecutive bytes making one binary integer. This volume is
	part of the Chapel Hill Volume Rendering Test Dataset, Volume I
	See the file `ChapelHill1.txt' for a description of how to get
	this excellent data set.
	
Format: A directory containing 127 images, each file with 256 rows,
	256 columns, 2 bytes/pixel

Acknowledge: 	Siemens Medical Systems, Inc., Iselin, NJ.  
        SoftLab, University of North Carolina, Chapel Hill, NC.

onion.ic
================================================
Origin: Computer Vision Laboratory, Link\"oping University, 
	Link\"oping, Sweden.

History: Synthetic 3D data designed for testing of 3D
	line/edge/orientation detectors, see:  

	@PhdThesis{b91,
	  author = 	{H. B{\aa}rman},
	  title = 	{Hierachical Curvature Estimation in Computer Vision},
	  school = 	{Link\"oping University, Sweden},
	  year = 	{1991},
	  address = 	{S--581 83 Link\"oping, Sweden},
	  month = 	{September},
	  note = 	{Dissertation No 253, ISBN 91--7870--797--8}
	}

	@PhdThesis{h92,
	  author = 	{L. Haglund},
	  title = 	{Adaptive {M}ultidimensional {F}iltering},
	  school = 	{Link\"oping University, Sweden},
	  year = 	{1992},
	  address = 	{S--581 83 Link\"oping, Sweden},
	  month = 	{October},
	}

	This volume contains all possible 3D plane orientations for a
	wide spatial frequency range. A traveller headed outwards from
	the centre and moving in a straight line would experience a
	sine wave with decreasing frequency, ie `onion' is a badly
	chosen name. 


Format: A directory containing 64 slices (images), each file with 64
	rows, 64 columns, 1 unsigned byte/pixel. 

Acknowledge: Hans Knutsson, Leif Haglund, Hakan Barman.

roadworkC.ic
================================================
Origin: L. Olsson, SAAB-SCANIA, Jonkoping, Sweden

History: A colour sequence. Video recorded from a SAAB 9000 using
	super-VHS. Transfered to the digital video standard CCIR 601.

Format: A directory containing 36 frames (108 images). Note that
	the sequence is stored in YUV format and NOT RGB! Each frame
	has three image files, one for each YUV-component. The
	digital video standard (CCIR  601)  defines  a  4:2:2
	sampling, ie  13.5 MHz for luminance and 6.5 MHz  for  each
	of  the  chrominance  components.  Because  of  this  the Y
	component image has double the width of the U and V
	components. 

	The Y (luminance) images are roadworkC_0.imf, roadworkC_3.imf,
	roadworkC_6.imf, etc with 576X720 pixels (1 unsigned byte)
	The U and V images have 576X360 1 unsigned byte/pixel.

	YUV => RGB conversion:
		R= 0.9709*Y - 0.0516*U + 1.4026*B
		G= 0.9709*Y - 0.3960*U - 0.7144*B
		B= 0.9709*Y + 1.7214*U - 0.0001*B

Acknowledge: SAAB-SCANIA.


tbrus3D.ic
================================================
Origin: Computer Vision Laboratory, Link\"oping University, 
	Link\"oping, Sweden.

History: Synthetic 3D white gaussian noise. Sdev=32.1
	@PhdThesis{b91,
	  author = 	{H. B{\aa}rman},
	  title = 	{Hierachical Curvature Estimation in Computer Vision},
	  school = 	{Link\"oping University, Sweden},
	  year = 	{1991},
	  address = 	{S--581 83 Link\"oping, Sweden},
	  month = 	{September},
	  note = 	{Dissertation No 253, ISBN 91--7870--797--8}
	}
	@PhdThesis{h92,
	  author = 	{L. Haglund},
	  title = 	{Adaptive {M}ultidimensional {F}iltering},
	  school = 	{Link\"oping University, Sweden},
	  year = 	{1992},
	  address = 	{S--581 83 Link\"oping, Sweden},
	  month = 	{October},
	}

Format: A directory containing 64 slices (images), each file with 64
	rows, 64 columns, 1 unsigned byte/pixel. 

Acknowledge: Hans Knutsson.




=====================================
##### Description of the images #####
=====================================

bredploop.imf
=====================================
Origin: Computer Vision Lab. Link\"oping University, Link\"oping, Sweden.

History: Designed for testing of line/edge/orientation detectors, see:
	@PhdThesis{h92,
	  author = 	{L. Haglund},
	  title = 	{Adaptive {M}ultidimensional {F}iltering},
	  school = 	{Link\"oping University, Sweden},
	  year = 	{1992},
	  address = 	{S--581 83 Link\"oping, Sweden},
	  month = 	{October},
	}
	This image covers a wider frequency range than ploop.imf (see below)

Format: 512 rows, 512 columns, 1 unsigned byte/pixel.

Acknowledge: Leif Haglund & Hans Knutsson

girl.imf
=====================================
Origin: The well-known `Lenna' image in BW. The grey level has been
	computed from the RGB images in the image data base of the USC
	Image Processing Institute.  

History: From Playboy November '74. (I will NOT post her address /hb).

Format: 512 rows, 512 columns, 1 unsigned byte/pixel.

NOTE!   This image is currently not available by ftp. We can use it
	ourselves but NOT distribute it (USCIPI report 1060).   


ploop.imf
=====================================
Origin: Computer Vision Lab. Link\"oping University, Link\"oping, Sweden.

History: Designed for testing of line/edge/orientation detectors, see:
		@phdthesis{k82,
		AUTHOR={Hans Knutsson},
		TITLE={Filtering and Reconstruction in Image Processing},
		SCHOOL={Link\"oping University, Sweden},
		YEAR={1982},
		NOTE={Diss. No. 88}}


Format: 512 rows, 512 columns, 1 unsigned byte/pixel.

Acknowledge: Hans Knutsson.

tbrus.imf
================================================
Origin: Computer Vision Laboratory, Link\"oping University, 
	Link\"oping, Sweden.

History: Synthetic 2D white gaussian noise. Sdev=32.1 (same data as
	 tbrus3D.ic!) 

Format: see ploop.imf

Acknowledge: see ploop.imf


-- 

-------------------------------------------------------------------------------
H}kan B}rman	                   | Email: hobo@isy.liu.se
Linkoping University		   |