Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Tell-and-Answer: Towards Explainable Visual Que...
Search
onizuka laboratory
December 18, 2018
Research
0
72
Tell-and-Answer: Towards Explainable Visual Question Answering using Attributes and Captions
弊研究室で行なったEMNLP2018読み会の発表資料です。
onizuka laboratory
December 18, 2018
Tweet
Share
More Decks by onizuka laboratory
See All by onizuka laboratory
Phrase-Based & Neural Unsupervised Machine Translation
onilab
0
120
Card-660: A Reliable Evaluation Framework for Rare Word Representation Models
onilab
0
36
A Word-Complexity Lexicon and A Neural Readability Ranking Model for Lexical Simplification
onilab
0
130
Integrating Transformer and Paraphrase Rules for Sentence Simplification
onilab
0
60
An Auto-Encoder Matching Model for Learning Utterance-Level Semantic Dependency in Dialogue Generation
onilab
0
56
Generating More Interesting Responses in Neural Conversation Models with Distributional Constraints
onilab
0
100
Modeling Multi-turn Conversation with Deep Utterance Aggregation
onilab
0
96
Learning Semantic Sentence Embeddings using Pair-wise Discriminator
onilab
0
120
SGM: Sequence Generation Model for Multi-Label Classification
onilab
0
80
Other Decks in Research
See All in Research
論文紹介: ReGenesis: LLMs can Grow into Reasoning Generalists via Self-Improvement
hisaokatsumi
0
110
A scalable, annual aboveground biomass product for monitoring carbon impacts of ecosystem restoration projects
satai
4
350
MetaEarth: A Generative Foundation Model for Global-Scale Remote Sensing Image Generation
satai
4
320
2021年度-基盤研究B-研究計画調書
trycycle
PRO
0
350
日本語新聞記事を用いた大規模言語モデルの暗記定量化 / LLMC2025
upura
0
240
電通総研の生成AI・エージェントの取り組みエンジニアリング業務向けAI活用事例紹介
isidaitc
1
1k
EarthDial: Turning Multi-sensory Earth Observations to Interactive Dialogues
satai
3
230
[RSJ25] Enhancing VLA Performance in Understanding and Executing Free-form Instructions via Visual Prompt-based Paraphrasing
keio_smilab
PRO
0
140
心理言語学の視点から再考する言語モデルの学習過程
chemical_tree
2
610
cvpaper.challenge 10年の軌跡 / cvpaper.challenge a decade-long journey
gatheluck
3
350
世界の人気アプリ100個を分析して見えたペイウォール設計の心得
akihiro_kokubo
PRO
62
31k
EcoWikiRS: Learning Ecological Representation of Satellite Images from Weak Supervision with Species Observation and Wikipedia
satai
3
260
Featured
See All Featured
Testing 201, or: Great Expectations
jmmastey
45
7.7k
Exploring the Power of Turbo Streams & Action Cable | RailsConf2023
kevinliebholz
34
6.1k
Optimizing for Happiness
mojombo
379
70k
Measuring & Analyzing Core Web Vitals
bluesmoon
9
620
Keith and Marios Guide to Fast Websites
keithpitt
411
23k
Facilitating Awesome Meetings
lara
56
6.6k
Rails Girls Zürich Keynote
gr2m
95
14k
Designing Experiences People Love
moore
142
24k
We Have a Design System, Now What?
morganepeng
53
7.8k
GitHub's CSS Performance
jonrohan
1032
470k
CoffeeScript is Beautiful & I Never Want to Write Plain JavaScript Again
sstephenson
162
15k
Into the Great Unknown - MozCon
thekraken
40
2.1k
Transcript
Tell-and-Answer: Towards Explainable Visual Question Answering using Attributes
and Captions Q. Li, J. Fu, D. Yu et al. EMNLP 2018 20181218
VQA CNN RNN ;* end-to-end !" → <2AE+H
C0 A9#5 2 4G' C0 <2I!".(AE)= >'1 %FI6?B,?:$D/-8 3@ VQA end-to-end &7 2 4G' <2AE)= 1
Visual Q&A Q: where is
the man swinging the racket? A: tennis court 2
Visual Q&A Q: what kind
of drink is in the glass? A: water 3
Visual Q&A Q: what is
walking next to the bus? A: cow 4
Visual Q&A Q: does the
man need a haircut? A: yes 5
Tell-and-Answer: Towards Explainable Visual Question Answering using Attributes and
Captions 6
"# where is the man swinging
the racket? yes no water tennis court ⋮ CNN RNN $ ! 7
7# 9+ D >(. <& D (CB D *
"6* D -2,4 end-to-end 2 3A $% @8181;?':/4-2 0),4 =C !581;?B 8
Tell-and-Answer: Towards Explainable Visual Question Answering using Attributes and
Captions 9
( ,# &+ Ø#'! Ø#(" ) ,( %*&$ ! Ø'
ver. Ø( ver. Ø ver. ( ) 10
11
@,→H3L0 !%$ %)=1M ;F$ % @, I87 H3L0 ResNet152
'#(' .? ØBAK6 ED "&'&( .?9 /$ % -NC> .? G+ ;F$ % cos N*2 J17 H3L0 cos N*2.? 5:4< H3)= 12
/ →3*) / ResNet152 LSTM 1 .%0
1', (e.g. BLEU) 4") .%) 2$5! cos 6# (+&- / 3*).% 13
7*='?8.-?9(-→64 #2(> 8.-9(- LSTM !" Ø LSTM &7!!)<%/
;5 $5') 3: softmax ,0+1 64#2 14
Tell-and-Answer: Towards Explainable Visual Question Answering using Attributes and
Captions 15
-VQA-real Ø1 & 3 ,1 10
)', *!#-min(#)*+,-. /010-/ 2),2 ,-.345 6 , 1) Ø%)' 10 " 3 ( $) + * 16
VQA vs.
17
vs.
18
VQA *!-# +$ .% -#"0,)( 0'/ Ø& -#" Ø
-#" ØNULL 1 -# $ 19
* '" Ø+! & )%
Ø+! & )% Ø+! & )% Ø+! & )% $ ,# - +! ( VQA & 20
tennis, ball, man, racket, hit, court, play, player,
swing, hold a man holding a tennis racket on a tennis court. tennis court & Q: where is the man swinging the racket? A: tennis court 21
bicycle, man, sit, eat, bike, look, outside, food,
person, table a man sitting at a table with a plate of food. beer & Q: what kind of drink is in the glass? A: water 22
street, bus, cow, city, walk, car, drive, stand,
road, white a cow that is walking in the street. car & Q: what is walking next to the bus? A: cow 23
woman, bear, teddy, hold, sit, glass, animal, large,
lady a woman holding a sandwich in her hands. yes & Q: does the man need a haircut? A: yes 24
30%
65% yes/no 80% 25
VQA 26
Tell-and-Answer: Towards Explainable Visual Question Answering using Attributes and
Captions 27
>1=9 5 2 3B# VQA #2 Ø7!6*A-"?&.% / Ø7!
<+; 0',4:(C $ 8 VQA =@ ) = 28