Project Overview It is important to recognize materials -- the stuff - - PowerPoint PPT Presentation

▶

Mar 01, 2023 202 likes •248 views

MIT2000-02: The Recognition of Material Properties for Vision Edward H. Adelson Project Overview It is important to recognize materials -- the stuff that objects are made of. Example: a domestic robot must distinguish between a

SLIDE 1

NTT - MIT Research Collaboration — Bi-Annual Report, July 1, 2002 — December 31, 2002

MIT2000-02: The Recognition of Material Properties for Vision

Edward H. Adelson

Project Overview

It is important to recognize materials -- the “stuff”

that objects are made of.

Example: a domestic robot must distinguish between

a pile of sugar and a blob of cream cheese in order to clean up properly.

Other example uses:

– Grasping an object: is it glass, metal, or rubber? – Locomotion over terrain: is it icy, or snowy, or wet, or sandy? – Mineralogy: how to classify minerals by their appearance? – Medicine: is this a melanoma or a normal mole?

SLIDE 2

NTT - MIT Research Collaboration — Bi-Annual Report, July 1, 2002 — December 31, 2002

MIT2000-02: The Recognition of Material Properties for Vision

Edward H. Adelson

Progress Through December 2002

Psychophysics of Reflectance Estimation

– Identified a number of properties of real-world illumination that are important for surface reflectance estimation (e.g. pixel histogram skew)

Input Image Shading Image Reflectance Image

Recovering Intrinsic Images from a Single Image

– Refined techniques for propagating information from reliable areas

f the image into ambiguous areas

Getting rid of histogram skew makes things look less glossy: Adding histogram skew makes things look more glossy:

SLIDE 3

NTT - MIT Research Collaboration — Bi-Annual Report, July 1, 2002 — December 31, 2002

MIT2000-02: The Recognition of Material Properties for Vision

Edward H. Adelson

Research Plan for the Next Six Months

Psychophysics of Reflectance Estimation

– Replicate and extend our experiments with surface shapes

ther than spheres.

– Perform experiments with surfaces that have empirically measured BRDFs, rather than parametric approximations.

Recovering Intrinsic Images from a Single

MIT2000-02: The Recognition of Material Properties for Vision

Project Overview

that objects are made of.

a pile of sugar and a blob of cream cheese in order to clean up properly.

– Grasping an object: is it glass, metal, or rubber? – Locomotion over terrain: is it icy, or snowy, or wet, or sandy? – Mineralogy: how to classify minerals by their appearance? – Medicine: is this a melanoma or a normal mole?

MIT2000-02: The Recognition of Material Properties for Vision

Progress Through December 2002

– Identified a number of properties of real-world illumination that are important for surface reflectance estimation (e.g. pixel histogram skew)

– Refined techniques for propagating information from reliable areas

MIT2000-02: The Recognition of Material Properties for Vision

Research Plan for the Next Six Months

– Replicate and extend our experiments with surface shapes

– Perform experiments with surfaces that have empirically measured BRDFs, rather than parametric approximations.

Image

– Investigate decomposing different characteristics of the scene – Extend technology to computer graphics applications