Data from "Context-Aware Textures"

Data from “Context-Aware Textures”

Data Description

This dataset was captured in the Graphics Lab at Yale University. We have captured spatially and temporally varying appearances due to weathering effects, and correlated these variations to geometry and environmental factors.

A ShapeGrabber laser triangulation scanner was used to scan the object shape. Depending on the size and complexity of the shape, 10 to 30 separate scans were collected for each model. The scan data was processed into a 3D shape using the scan processing system distributed by the Visual Computing Laboratory of ISTI-CNR [Callieri et al. 2002], including scan data cleaning, aligning and merging, hole-filling, and simplifying. Based on the mesh model, shape-related contexts were estimated either directly (like surface curvature and principle directions), by digital simulations (like ambient occlusion and source direction), or with the help from optical devices (like coating thickness).

An Olympus C8080WZ color camera was calibrated with respect to the ShapeGrabber coordinate system (a description of the design and calibration of this system is given in [Farouk et al. 2003]). It then captured images in each view of the object under lighting from five small, halogen lights, whose positions had been measured in the scanner coordinate system. The color images were processed to eliminate the effects of illumination direction and to produce maps of the diffuse component of surface reflectance using methods described in [Bernardini et al. 2001]. Although our samples are not all Lambertian, the diffuse texture maps acted as maps of where change had occurred on the surface. A texture map was computed for the entire object. A complete shape scan was subsequently performed at each time step, so that by aligning the scanned shapes the texture maps for all time steps were aligned and changes can be tracked at each texture-map pixel. (Please refer to original paper for paper references.)

Data Format

For each weathering effect we studied, both source and target objects are captured but shared in slightly different formats:

Source Objects: Shape is shared as a VRML model; time-varying appearances are attached as multiple texture atlases for that VRML model; individual context measurements are stretched to the range of [0,255] and rendered aligned with those texture atlases, with invalid regions marked out using green color (#00ff00).
Target Objects: Shape is shared as a PLY file and an IV file. Both of them contain 3D vertices coordinates and vertices indices for each triangle, in the same order, in ASCII mode. All context parameters files begin with vertices number, followed by values corresponding to each vertex in PLY or IV file, in the same order.

Data Repository

The following table summarizes all shared data. Please refer to the original paper for the conduction of each weathering effect, and definition of each context parameter. Please see section of “Contact for Downloading Data” for downloading instruction.

Figure	effect	source model	target model(s)	context parameters
Figure	effect	source model	target model(s)	occlu.	pdir.	curv.	sdir.	thick.
12	patina	CopperMold	Seahorse; Fairy	+
13	rusting(b)	IronMozart	Urn; CandleStick	+			+
14(left)	rusting(a)	IronRipple	IronPot	+			+
14(right)	rusting(c)	BlackironTjunction	BlackironElbow	+			+
15	mold growing	CheeseBach	StringCheese; SwissCheese	+
16	paint crackling	Slide	Frog; SippyLion					+

Fig. 12, Patina

CopperMold (as source): [Note: only image series aligned with ambient occlusion are provided; context values within [0,1] are stretched to [0,255] in the TIF image.]
Seahorse (as target): 185,681 vertices, 365,421 triangles
Fairy (as target): 239,000 vertices, 478,004 triangles

Fig.13, Rusting(b)

IronMozart (as resouce): 115,621 vertices, 217,484 triangles [Note: contexts of ambient occlusion and source direction are combined into one single context parameter, as “(1-ambient_occlusion) * source_direction”; context values within [0,1] are stretched to [0,255] in the TIF image.]
Urn (as target): 140,404 vertices, 280,281 triangles [Note: context of source direction is “1” for all vertices; this indicates some “ambient” source; user can simply ignore this context in texture synthesis.]
CandleStick (as target): 167,417 vertices, 334,736 triangles [Note: see source data for definition of single context.]

Fig.14(left), Rusting(a)

IronRipple (as source): 253,767 vertices, 479,236 triangles [Note: context values within [0,1] are stretched to [0,255] in the TIF image.]
IronPot (as target): 105,565 vertices, 211,018 triangles [Note: context of source direction is “1” for all vertices - this indicates some “ambient” source; users can simply ignore this context in texture synthesis, or specify their own source direction.]

Fig.14(right), Rusting(c)

BlackironTjunction (as source): 47,037 vertices, 85,221 triangles [Note: contexts of ambient occlusion and source direction are combined into one single context parameter, as “(1-ambient_occlusion) * source_direction”; context values within [-1,1] are stretched to [0,255] in the TIF image.]
BlackironElbow (as target): 36,783 vertices, 67,176 triangles [Note: see source data for definition of single context.]

Fig.15, Mold Growing

CheeseBach (as source): 68,167 vertices, 126,661 triangles [Note: context values within [0,1] are stretched to [0,255] in the TIF image.]
StringCheese (as target): 61,767 vertices, 123,530 triangles
SwissCheese (as target): 34,102 vertices, 67,698 triangles

Fig.16, Paint Crackling

Slide (as source): [Note: source object is a 2D transparency slide, therefore, only image series aligned with coating thickness are provided; context values within [0,1] are stretched to [0,255] in the TIF image.]
Frog (as target): 160,050 vertices, 319,162 triangles
SippyLion (as target): 6,518 vertices, 12,000 triangles

Copyright Declaration

Anyone is free to use this dataset for research purposes, only. Without permission from Yale University, this data cannot be incorporated into a larger database, which is then publicly distributed. If experimental results are obtained based on the data here, all publications of these results should acknowledge its use by referencing the following paper:

@article{Lu-07-ContextAwareTextures,

author = { Jianye Lu and

Athinodoros S. Georghiades and

Andreas Glaser and

Hongzhi Wu and

Li-Yi Wei and

Baining Guo and

Julie Dorsey and

Holly Rushmeier },

title = { Context-Aware Textures },

journal = { ACM Trans. Graph. },

volume = { 26 },

number = { 1 },

year = { 2007 },

publisher = {ACM Press},

}

Contact for Downloading Data

All these data are shared via ftp://wanda.cs.yale.edu/. Please send email to “jianye.lu AT yale.edu” with subject starting with “[CAT]” using your institution or company email account; user name and password will be sent to you at no charge. In return, we would like to know who you are and which institution or company you work for in your email.

This page has been tested in Firefox1.5 / IE7.0 on WindowsXP, and Firefox1.5 / Opera9.0 / Konquerer3.5 on Linux. If your web browser cannot visit the FTP site directly, please use other ftp clients.

If you have any question or comment about this page, please also send email to “jianye.lu AT yale.edu” with subject starting with “[CAT]”.

Last Updated: June 18, 2007