Learning from Descriptive TextTamara L Berg
Stony Brook University
Vision
Humans
Language
Tags: canon, eos, macro, japan, vacation, frog, animal, toad, amphibian, pet, eye, feet, mouth, finger, hand, prince, photo, art, light, photo, flickr, blurry, favorite, nice.
It's the perfect party dress. With distinctly feminine details such as a wide sash bow around an empire waist and a deep scoopneck, this linen dress will keep you comfortable and feeling elegant all evening long.
Visually Descriptive Text
Visually descriptive language provides:• information about how people construct natural language for
imagery.• information about the world, especially the visual world.• guidance for computational visual recognition.
How do peopledescribe the world?
“It was an arresting face, pointed of chin, square of jaw. Her eyes were pale green without a touch of hazel, starred with bristly black lashes and slightly tilted at the ends. Above them, her thick black brows slanted upward, cutting a startling oblique line in her magnolia-white skin–that skin so prized by Southern women and so carefully guarded with bonnets, veils and mittens against hot Georgia suns” –
Gone with the Wind
How does theworld work?
What should we recognize?
Visually Descriptive Text
Visually descriptive language provides:• information about how people construct natural language for
imagery.• information about the world, especially the visual world.• guidance for computational visual recognition.
How do peopledescribe the world?
“It was an arresting face, pointed of chin, square of jaw. Her eyes were pale green without a touch of hazel, starred with bristly black lashes and slightly tilted at the ends. Above them, her thick black brows slanted upward, cutting a startling oblique line in her magnolia-white skin–that skin so prized by Southern women and so carefully guarded with bonnets, veils and mittens against hot Georgia suns” –
from Gone with the Wind by Margaret Mitchell
How does theworld work?
What should we recognize?
What’s in a description?
What do people describe?“A bearded man is holding a child in a sling.”
manbabyslingshirtglassesladderfridgetablewatermelonchairboxescupswater bottlewallpacifierbeard…
“A bearded man stands while holding a small child in a green sheet.” “A bearded man with a baby in a sling poses.”“Man standing in kitchen with little girl in green sack.” “Man with beard and baby”
What’s in this image?
What’s in a description?
Predict what people will describeGiven an image
“looking for castles in the clouds out my car window”
1)
2)
Given a caption
“two women sitting brunette blonde on bench reading magazine”
Predict what’s in the image
clouds ✔car
window
castle
✖✖?
women ✔bench ✔
magazine ✔grass
skirt✖✖…
e.g. Spain & Perona, 2010
President George W. Bush makes a statement in the Rose Garden while Secretary of Defense Donald Rumsfeld looks on, July 23, 2003. Rumsfeld said the United States would release graphic photographs of the dead sons of Saddam Hussein to prove they were killed by American troops. Photo by Larry Downing/Reuters
Who’s in the picture?T.L. Berg, A.C. Berg, J. Edwards, D.A. Forsyth
Model Accuracy of labelingVision model, No Lang model
67%
Vision model + Lang model 78%
Visually Descriptive Text
Visually descriptive language provides:• information about how people construct natural language for
imagery.• information about the world, especially the visual world.• guidance for computational visual recognition.
How do peopledescribe the world?
“It was an arresting face, pointed of chin, square of jaw. Her eyes were pale green without a touch of hazel, starred with bristly black lashes and slightly tilted at the ends. Above them, her thick black brows slanted upward, cutting a startling oblique line in her magnolia-white skin–that skin so prized by Southern women and so carefully guarded with bonnets, veils and mittens against hot Georgia suns” –
from Gone with the Wind by Margaret Mitchell
How does theworld work?
What should we recognize?
Vision is hard
World knowledge (from descriptive text) can be used to smooth noisy vision predictions!
Green sheep
Learning World Knowledge
Attributes
Relationships
green green grass by the lake
a very shiny car in the car museum in my hometown of upstate NY.
Our cat Tusik sleeping on the sofa near a hot radiator.
very little person in a big rocking chair
BabyTalk: Understanding and Generating Simple Image DescriptionsKulkarni, Premraj, Dhar, Li, Choi, AC Berg, TL Berg, CVPR 2011
System Flow
Input Image
Extract Objects/stuff
a) dog
b) person
c) sofa
brown 0.32striped 0.09furry .04wooden .2Feathered .04 ...
brown 0.94striped 0.10furry .06wooden .8Feathered .08 ...
brown 0.01striped 0.16furry .26wooden .2feathered .06 ...
a) dog
b) person
c) sofaPredict attributesPredict prepositions
a) dog
b) person
c) sofa
near(a,b) 1 near(b,a) 1 against(a,b) .11against(b,a) .04 beside(a,b) .24beside(b,a) .17 ...near(a,c) 1 near(c,a) 1 against(a,c) .3against(c,a) .05 beside(a,c) .5beside(c,a) .45 ...near(b,c) 1 near(c,b) 1 against(b,c) .67against(c,b) .33 beside(b,c) .0beside(c,b) .19 ...
Predict labeling – vision potentials smoothed with text potentials
<<null,person_b>,against,<brown,sofa_c>> <<null,dog_a>,near,<null,person_b>> <<null,dog_a>,beside,<brown,sofa_c>> Generate natural
language description
This is a photograph of one person and one brown sofa and one dog. The person is against the brown sofa. And the dog is near the person, and beside the brown sofa.
This is a picture of one sky, one road and one sheep. The gray sky is over the gray road. The gray sheep is by the gray road.
Here we see one road, one sky and one bicycle. The road is near the blue sky, and near the colorful bicycle. The colorful bicycle is within the blue sky.
BabyTalk results
This is a picture of two dogs. The first dog is near the second furry dog.
Objects, Attributes, Prepositions
Visually Descriptive Text
Visually descriptive language provides:• information about how people construct natural language for
imagery.• information about the world, especially the visual world.• guidance for computational visual recognition.
How do peopledescribe the world?
“It was an arresting face, pointed of chin, square of jaw. Her eyes were pale green without a touch of hazel, starred with bristly black lashes and slightly tilted at the ends. Above them, her thick black brows slanted upward, cutting a startling oblique line in her magnolia-white skin–that skin so prized by Southern women and so carefully guarded with bonnets, veils and mittens against hot Georgia suns” –
from Gone with the Wind by Margaret Mitchell
How does theworld work?
What should we recognize?
What should we recognize?
• Recognition is beginning to work
• Open question – what should we recognize?
• Maybe objects aren’t (always) the right base level entities
Object Recognition
Parts, Poselets, AttributesFor example: [Fergus, Perona, Zisserman2003],[Bourdev, Malik2009], …
Slide Credit: Ali Farhadi
Learn which terms in descriptions are depictable
Fully beaded with megawatt crystals, this Christian Louboutin suede pump matches the gleam in your eye.
Pump's linear heel plays up the alluring curves of its dipped sides.
Round toe frames low-cut vamp.Tonally topstitched collar.
4" straight, covered heel shows off signature red sole.
Creamy leather lining with padded insole."Fifi" is made in Italy.
attributes
Automatically Discovering Attributes from Noisy Web Data T.L. Berg, A.C. Berg, J. Shih ECCV 2010
Given Web Images + Noisy Text Descriptions: 1) Discover visual attribute terms in text descriptions - likely domain dependent 2) Learn appearance models for attributes without labeled data 3) Characterize attributes by: type, localizability
Object Recognition
For example: [Oliva, Torralba 2001],[SUN 2010], …
Slide Credit: Ali Farhadi
Scenes
What are the right quanta of Recognition?
Farhadi & SadeghiRecognition using Visual Phrases , CVPR 2011
Participating in Phrases Profoundly affects the appearance of objects
Farhadi & SadeghiRecognition using Visual Phrases , CVPR 2011
Maybe descriptive text can inform entity hypotheses!
“the dog is sleeping”
“sleeping dog in delhi”“A dog is sleeping in”
“a sleeping dog in NTHU”
What should we recognize?
“cat in a bag”
“cat in the bag”“cat in bag”
“the cat is in the bag”
What should we recognize?
Maybe descriptive text can inform entity hypotheses!
Conclusion
Use large pools of descriptive text to: Learn how people describe the visual world Learn how the world works Guide future efforts in recognition
Apply this knowledge to multi-modal
collections & applications
Acknowledgements
• Collaborators: Alex Berg, David Forsyth, Jaety Edwards, Jonathan Shih, Girish Kulkarni, Visruth Premraj, Sagnik Dhar, Vicente Ordonez, Siming Li, Yejin Choi, Kota Yamaguchi, Vicente Ordonez
• Funded by NSF Faculty Early Career Development (CAREER) Program: Award #1054133