计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

78
计计计计计计 计计计计计计 Computer Seeing and Hearing-A Dream of AI 计计 计计计计计计计计计计计计计 计计计计计计计计计计计计计 计计计计计计计计计计计计计计 计计计计计计 计计计计计计计

description

计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI. 张钹 清华大学信息科学与技术学院 清华大学计算机科学与技术系 清华信息科学与技术国家实验室 智能技术与系统国家重点实验室. Computer Vision /Hearing. Is it possible ? Yes No It is just a daydream !. The Characteristic of Auditory Information (Data). Ears, Earphones - PowerPoint PPT Presentation

Transcript of 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Page 1: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

计算机视听觉-人工智能之梦Computer Seeing and Hearing-A

Dream of AI

张钹清华大学信息科学与技术学院清华大学计算机科学与技术系清华信息科学与技术国家实验室智能技术与系统国家重点实验室

Page 2: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Is it possible ?YesNo It is just a daydream !

Computer Vision /Hearing

Page 3: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

The Characteristic of Auditory Information (Data)

Ears, Earphones A continuous waveDigital Data: 20K-100K bits/sSparseness (Redundant)Noisy

Page 4: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

The Characteristics of Visual Information (Data)

Eyes, Digital Camera • Pixel-based (million, ten million bits) Sparseness (Redundancy) Noisy• Eyes: a sequence of images 109 bits/sec

Page 5: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

The Sparseness of Auditory Signal

采样频率 位分辨率• 广播质量- 48kHz• CD 质量- 44kHz 16 位• 收音音质- 22kHZ 8 位• 可接受的音乐- 11kHz 4 位• 可接受的语音- 5kHz

Page 6: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

The Sparseness of Visual Signal

分辨率与识别率的关系 (conceptual)

Page 7: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

一个不适定问题An Ill-posed Problem

Sparse, redundant, noisy data(110000111100011100011000………… )

Microphone (Ears)(Camera (Eyes))

Speaker-invariant Vowel RepresentationVowel-invariant Speaker Representation( Object-invariant Representation )

Existence Uniqueness Stability

Page 8: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

1. Segmentation & Recognition

Page 9: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Image Segmentation vs. Recognition

Which comes first, Chicken or Egg

Where is the object ?

What is the object ?

?

Page 10: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Speech Segmentation vs. Recognition

? What, Where

Page 11: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

技术上的困难(Technology)

Sparse, redundant, noisy data

Speaker-invariant Vowel RepresentationVowel-invariant Speaker Representation

A Robust Detector

An Invariant Descriptor

Page 12: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Top-down feedback

Top-down feedback

Local connection

Data-driven From egg to chicken

High-levelApriori-knowledge

人类是如何解决的?

Page 13: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

The Relation Between Activation Patterns and Early Stages of Sound Processing

Speech Encoding occurs not only in specialized high-level region but also in early stages of sound processing. The early sound processing may exhibit complex spectrotemporal receptive fields and may participate in high-level encoding of auditory objects, e.g., via local feedback

Page 14: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Multi-layer Neural Network with feedback connections

G. E. Hinton, The “wake-sleep” algorithms for unsupervised neural networks, SCIENCE vol.268, 26 May 1995, 1158-1161

Page 15: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

RepresentationRBM:Restricted Boltzmann Machine

Page 16: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Experimental Results

G. E. Hinton, Learning multiple layers of representation, TRENDS inCognitive Sciences vol.11, no.10, 428-434, 2007

Page 17: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

2 、 Feature Extraction

Page 18: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Computer Robustly Extractable Features

Sparse, redundant, noisy dataStatistical

Approaches

Speech-base Invariant Statistics (Features)

Speaker-invariant Vowel RepresentationVowel-invariant Speaker Representation

Page 19: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Statistical Method• 选择一个语音训练库• 提取语音特征• 无监督学习( Classification )• 分类准则- Generalization 提取何种特征 ?Computer robustly detectable

Page 20: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Representation at Different Granularities

Global Features-one vector The coarsest

The finest

Pixel Based-1280X800X3 vectors

An Image

Page 21: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Pixel-based Representation-the finest representation

• • • • • • • • • • •• • • • • • • • • • •• • • • • • • • • • •• • • • • • • • • • •• • • • • • • • • • •• • • • • • • • • • •• • • • • • • • • • •• • • • • • • • • • •• millionX3-dimensional vectors -all the details , ,

( , )

( , ) [ ( , )], 1 ,i j

k i j

X F f

X x y

F f g x y i j n

( , ) ,1 ,k kG g i j i j n

Page 22: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Global Features -the coarsest representation

N

jiji P

Nu

1

121

1

2 ))(1(

N

jiiji uP

N

Color moments

31

1

3 ))(1(

N

jiiji uP

Ns

N-the number of pixels, P-the value of each colorOne 9-dimensional vector

Page 23: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Coarse vs. Fine Representation

Representations

The Finest Representation

The Coarsest Representation

Expressiveness

Full Structural KnowledgeGood

No Structural KnowledgePoor

Robustness Poor Good (rotation, translation, scaling,…)

Page 24: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Representation with Middle Grain-Size

• • • • • • • • • • • • • • • •

• • • • • •

Region-based Representation

1 2

([ ] ,[ ] ,[ ] )

[ ] , 1, 2,...,

[ ] ( ), ( ),..., ( )k

i i i

i i k

i k k k n

X F f

X x i n

f f x f x f x

Page 25: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Local (Spatial) Feature Region-01 Region-11 Region-12

Foreground vs. Background

Page 26: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Vector Representation

1 2 1

[ ] :

( ), ( ),..., ( ) , 1,2,...,k

i

k k k n

f

f x f x f x k l

A set of vectors (tens) (with different length)Similarity MeasureWeighted

Page 27: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Region-adaptive Grid Partition

Jinhui Yuan (2005…)

Page 28: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Hierarchical (粒度)结构(X, F, f )-the finest space([X], [F], [f] )-coarse space[X] the quotient space of X[F] the quotient structure of F an equivalence class[f]-the quotient attributes of f

• • •

• • • • • • • • • • • • •

•••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••••

•Semantics (text, image)

Primitive (words, pixels)

Semantic Gap

Page 29: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

PM: Pyramid Match (feature space-quantization level) SPM: Spatial Pyramid Match (physical space-grid)FESCO: Feature Spatial Covariant Kernel

Page 30: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Concept Detection from Video Shots

Page 31: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

ExperimentsTRECVID 2005, 10 concepts 170 hours news (MSNBC, NBC Nightly News, CNN, LBC, CCTV, NTDTV)TRECVID 2006, 20 concepts 170+150 hours newsKeypoint descriptor: 64-dimensional SURF feature (Speeded Up Robust Features)AP: Non-interpolated Average PrecisionMAP: Mean Average Precision (7 concepts)

Page 32: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

TRECVID Data

Name hours no. shots no. frames dateTRECVID05d 80 44,000 75,000 2004 10-11TRECVID05t 80 46,000 78,000 2004 11-12TRECVID06t 150 80,000 144,000 2005 11-12TRECVID07d 50 18,000 18,000TRECVID07t 50 22,000 63,000

d: training data, t: testing data

Page 33: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Coarse vs. Fine Granulation

MAP: 7 concepts: car, explosion-fire, flag-US, maps, mountain, sports, waterscape-waterfront Test Set TRECV05t TRECV06tVocabulary Size

18 72 288 18 72 288

Grid 11 Grid 2 2 Grid 4 4

0.073 0.210 0.2440.223 0.260 0.2510.271 0.254 0.275

0.025 0.074 0.109 0.078 0.117 0.119 0.116 0.123 0.128

Page 34: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Multi-granulation

Combination

TRECV05t

TRECV06t

Whole Comb.

9 combinations 0.307 0.166

FESCO Fine SpaceFine FeatureFine Comb.Coarse Comb.

PiQj=288 Qj=G44; Pi=288, 72, 18 Pi=288; Qj=G11, G22, G44

PiQj>288 PiQj<288

0.306 0.300 0.294 0.293 0.250

0.166 0.158 0.155 0.151 0.106

MAP: 7 concepts: car, explosion-fire, flag-US, maps, mountain, sports, waterscape-waterfront

Page 35: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Multi-granulation (2)MAP: 7 concepts: car, explosion-fire, flag-US, maps, mountain, sports, waterscape-waterfront

Test Set TRECV05t TRECV06t Fusion Method

pre-fusion post-fusion

pre-fusion post-fusion

FESCO SPM PM

0.297 0.306 0.274 0.285 0.254 0.269

0.154 0.166 0.140 0.146 0.124 0.125

Page 36: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Multi-Granular & Multi-modalTRECVID2005 (Video Retrieval Evaluation Conference)86.6 hours of news videos (45766 shots in 140 video clips)Features: A: auto-speech recognition text T: visual texture R: color of segmented image regions

Page 37: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

PMSRA

Probabilistic Model Supported Rank Aggregation

Page 38: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

The Comparison between Uni-modal and Multi-granular, modal

Uni-Modal Multi-Granular, Modal

ASR Texture Region A+T A+R T+R A+T+R

US-flag 0.0335 0.0155 0.0375 0.0359 0.0506 0.0372 0.0521

Water 0.0034 0.1143 0.0814 0.1022 0.0735 0.1333 0.1211

Mountain 0.0033 0.0693 0.1104 0.0668 0.1066 0.1176 0.1154

Sports 0.0723 0.0769 0.2156 0.1465 0.2678 0.2802 0.3050

Average 0.0281 0.0690 0.1112 0.0879 0.1246 0.1421 0.1484

Page 39: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

TRECVID Text Retrieval Conference Video Retrieval Evaluation

Page 40: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

声波、声谱图( Spectrograms )

Page 41: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

语音信息Global Features-one vector The coarsest

The Finest-sampling

Page 42: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

不同粒度的语音特征• 语音单元(粒度)选择: 音素、音节、词… .• 语音参数选择 MFCC: Mel 频率倒谱参数 (Mel Frequency Cepstral Coefficients) LSP :线谱对 (Line Spectrum Pair) ICA (Independent Component Analysis)

• 多(粒度)特征融合

Page 43: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

3 、 Structural Model• Temporal Model (HMM)• Spatial Model

Page 44: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

语音的时间结构 (Temporal Structure)

多粒度结构

Page 45: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Image Region Annotation -horse, sky, mountain, grass, tree

Page 46: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Region-adaptive Grid Partition (2)

Page 47: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Experiments• 4002 Corel images (384256 or 256384)• 11 basic (region) concepts• Features: color moment + wavelet• 5 models: 2 without structural knowledge (GMM, SVM) 3 with structural knowledge (HMM*, RMF*, CRF*)

Page 48: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Image Region Annotation

Page 49: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Image Region Annotation

Page 50: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Spatial Structural Representation

n images, each image has mi=HV grids

( , ) ( , ), 1,2...,

( , ) ( , ), 1,2,...,i i

j ji i i i i

x y x y i n

x y x y j m

(a) i.i.d generative model(b) i.i.d. discriminative model(c) 2-dimensional hidden Markov (2D HMM)(d) Markov Random Field (MRF)(e) Conditional Random Field (CRF)

Page 51: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Different Models

Page 52: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Label Configuration ( , ), 1,2,...,i ix y i N

Given a training data, MAP (maximal a posterior) : label configuration

1: 1:* argmax ( )m my P y x

For 2D HMM, MRF, CRF using path limited Viterbi algorithm

Page 53: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Probabilistic distribution P Cs: labeling clique, C0: labeling and feature cliquey* the optimal label configuration

0

1: 1:

( , ) ( , )

1( , ) ( , ) ( , )i j s k k

m m i j k k

y y C y x C

P x y y y y xZ

1:0( , ) ( , )

* argmax ( , ) ( , )m

i j s k k

i j k k

yy y C y x C

y y y y x

Markov Random Field Model - MRF model

Page 54: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Comparison Among Different Models

GMM: Gaussian Mixture Model (30 components)SVM: Support Vector Machine Gaussian kennel, one-against-oneHMM: Hidden Markov ModelRMF: Random Markov FieldCRF: Conditional Random Field Limited Path Viterbi Algorithm

Page 55: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Experimental Results

Page 56: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

The Spatial Relation Among Region Labels

The probability that some things are above the “sky”, “flower” or “building”

Page 57: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Future Direction4. Data Driven Approach

Page 58: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

数据驱动法( Data-driven )数据驱动法的本质: 针对特定数据(语音、图像)库 高维空间的划分问题今后的发展方向:• Large scale annotated database• Sparseness in high dimensional space

Page 59: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

*******

Data Space

Page 60: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

HorsePrecision: 25/30 pictures

Global ColorFeatureHorse-Green

Page 61: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

EaglePrecision:13/25 pictures

Global Color FeatureEagle-Blue

Page 62: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

 

Local Features 17/36 picturesRegion-based Color Features Foreground Color pink

Page 63: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

The Bless of Dimensionality

Sparse RepresentationSample Space(Data Space)Extended Yale B2414 frontal-facewith different lighting38 individuals192168 image

J. Wright, et al. Robust face recognition via sparse representation, IEEE PAMI 08

Page 64: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Anti-Noise

30%

50%

70%

Page 65: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Anti-Occluded

Page 66: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

5. Brain Science (Structural Model)

Page 67: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

From eye to primary visual cortex

Li Zhaoping, Theoretical understanding of the early visual processes by data compression and data selection, Network: Computation in Neural Systems, December 2006; 17: 301-334

Page 68: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Two Basic Problems• Description: What is the object-invariant descriptor in human brain?• Detection: How to obtain the descriptor from a huge amount of data?There is some answer but is not a full answer.

Page 69: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Vision: 2D image- 3D scene

This is a hard problem even for human being

eyes + brain• Billions years evolution• 1/3 of brain resource• Several years learningMany problems are still unsolved for human

being

Page 70: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

基于人类认知的图像处理数据空间 感知空间(语义)

数据空间 原空间 语义空间2,000 bytes-50% (6464) 维特征 几十 bytes

Page 71: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Cognition (Perception) SpacePerception spaceSemantically meaningful features• 多层次 (hierarchy)• 自底向上的数据驱动 + 自顶向下的反馈(上下文,先验, 标注知识)

Page 72: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Object Recognitionwith sparse, localizedfeatures

MIT-CSAIL-TR-2006-028 T. Serre

Page 73: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI
Page 74: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

HMAX-sum + max

Page 75: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Computational Model

Page 76: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

Experimental Results• Caltech 101 The number of categories: 101 Training samples: 30/per class Average recognition rate: 51%• Vista (car, passerby, bicycle) AUC>90% • AUC: the area under the ROC (Receiver Operating Characteristics) curve

Page 77: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

人脑听觉皮层的试验研究Three Dutch vowels (a, i, u)Three speakers (1 female, 2 males)Features: F1-F2 F0

Elia Formisano “Who” Is Saying “What”? Brain-based Decoding of Human Voice and SpeechScience vol 322, 7 Nov. 2008

Page 78: 计算机视听觉-人工智能之梦 Computer Seeing and Hearing-A Dream of AI

谢谢 !