101 comp. graphics recent advances in cg...

12
CGT 101 Recent Advances in CG Bedrich Benes, Ph.D. Purdue University Department of Computer Graphics Technology and Department of Computer Science © Bedrich Benes CG and Other Fields Comp. Graphics describes the world visually Aims at perceived visual quality (not ONLY realism, also art) Computer vision attempts to reconstruct world (world computer model) Image processing works with images © Bedrich Benes World Description CG and Other Fields Geometry Image Light Materials Textures Materials Textures Rendering Rendering Modeling Modeling Lighting Lighting Image Processing Image Processing Computer Graphics Computer Graphics Reconstruction and Computer Vision © Bedrich Benes CG and Other Fields Examples: Computer Vision: 3D geometry reconstruction from images Texture capture Scanning 3D world Image understanding

Upload: others

Post on 23-Jul-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: 101 Comp. Graphics Recent Advances in CG ONLYhpcg.purdue.edu/bbenes/classes/CGT101/lectures/CGT101-06-Recen… · CGT 101 Recent Advances in CG Bedrich Benes, Ph.D. Purdue University

CGT 101Recent Advances in CGBedrich Benes, Ph.D.Purdue UniversityDepartment of Computer Graphics Technology andDepartment of Computer Science

© Bedrich Benes

CG and Other Fields• Comp. Graphics describes the world visually• Aims at perceived visual quality

(not ONLY realism, also art)

• Computer vision attempts to reconstruct world  (world computer model)

• Image processing works with images

© Bedrich Benes

World Description

CG and Other Fields

Geometry

ImageLight

MaterialsTexturesMaterialsTextures

RenderingRendering

ModelingModeling

LightingLighting

Image Processing

Image Processing

Computer GraphicsComputer Graphics

Reconstruction and Computer Vision © Bedrich Benes

CG and Other FieldsExamples: • Computer Vision: 

• 3D geometry reconstruction from images• Texture capture• Scanning 3D world• Image understanding

Page 2: 101 Comp. Graphics Recent Advances in CG ONLYhpcg.purdue.edu/bbenes/classes/CGT101/lectures/CGT101-06-Recen… · CGT 101 Recent Advances in CG Bedrich Benes, Ph.D. Purdue University

© Bedrich Benes

CG and Other FieldsExamples: • Image processing

• Image compression • Video codecs• Image inpainting• Computational cameras

© Bedrich Benes

Recent Advances

• Virtual Reality• Augmented Reality• Non‐Photorealistic Rendering• Artificial Intelligence

© Bedrich Benes

Virtual Reality

© Bedrich Benes

Scene

The Rendering ProcessWorld Description - Scene

Geometry

Light

MaterialsTextures

ModelingModeling

LightingLighting

World Description - Scene

Geometry

Light

MaterialsTextures

Modeling

Lighting

ImageCamera model

Page 3: 101 Comp. Graphics Recent Advances in CG ONLYhpcg.purdue.edu/bbenes/classes/CGT101/lectures/CGT101-06-Recen… · CGT 101 Recent Advances in CG Bedrich Benes, Ph.D. Purdue University

© Bedrich Benes

Virtual RealityLeft image

Right image

Scene

Left camera

Right camera

Left eye

Right eye

© Bedrich Benes

Virtual Reality• The rendering process is doubled• There are two cameras

approx. distance of the human eyes apart• They render two images• The images are displayed to left and right eye• The human brain perceives it as 3D• The viewers is immersed 

into the virtual environment

© Bedrich Benes

Virtual Reality• It needs precise head tracking

• Head movement is transferred to the dual‐camera movement

• The environment is simulatedall is synthetic

© Bedrich Benes

Virtual Reality• How do we deliver the data to each eye?• …by using special displays

• Shutter Glasses

• Head Mounted Displays (HMD)

Page 4: 101 Comp. Graphics Recent Advances in CG ONLYhpcg.purdue.edu/bbenes/classes/CGT101/lectures/CGT101-06-Recen… · CGT 101 Recent Advances in CG Bedrich Benes, Ph.D. Purdue University

© Bedrich Benes

Shutter Glasses• They turn on and off the display for 

each eye at high frequency• A specialized display displays 

the left and the right images in fast sequence

• It needs to be synchronized

https://en.m.wikipedia.org/wiki/Active_shutter_3D_system

© Bedrich Benes

Shutter Glasses• Nvidia 3D Vision (2008)

• USB‐support• Stereoscopic gaming kit

© NVidia

© Bedrich Benes

Shutter Glasses• The Cave

• Four large active displays• Active shutter glasses• Can move the screen walls

L, U, Flat

© Visibox Inc

© Purdue University

© Bedrich Benes

Shutter Glasses• Challenges

• Limited view (the screen)

• Limited motion

• Limited field of view

Page 5: 101 Comp. Graphics Recent Advances in CG ONLYhpcg.purdue.edu/bbenes/classes/CGT101/lectures/CGT101-06-Recen… · CGT 101 Recent Advances in CG Bedrich Benes, Ph.D. Purdue University

© Bedrich Benes

Head Mounted Displays• Double display • Worn on head• Precise head tracking• Good sound• Some support eye‐tracking

© Oculus

© Bedrich Benes

Head Mounted Displays• Challenges

• Motion sickness

• Heavy (it is getting better, but…)

• Needs huge computational power

© Bedrich Benes

VR• VR Challenges in General

• Lag • There is a delay between the movement and display

• Limited field of view• We perceive in the corners of our eyes too

• Full immersion is only visual• You cannot feel the weight of the objects.

• And all the imprecisions • Colors, sound, tracking… 

© Bedrich Benes

Augmented Reality

Page 6: 101 Comp. Graphics Recent Advances in CG ONLYhpcg.purdue.edu/bbenes/classes/CGT101/lectures/CGT101-06-Recen… · CGT 101 Recent Advances in CG Bedrich Benes, Ph.D. Purdue University

© Bedrich Benes

Augmented Reality• AR displays additional data in real world

• Usually an image in the real world

• Provides some additional insight

• A popular ones are real‐time translators

© Bedrich Benes

Augmented Reality

© Bentley https://www.youtube.com/watch?v=5HV3fcTvZk0

© Bedrich Benes

Augmented Reality

© Pepsi https://www.youtube.com/watch?v=Go9rf9GmYpM © Bedrich Benes

Augmented Reality

© Microsoft

• Technology• Head mounted 

transparent displayse.g., Microsoft HoloLens

• Cellphones 

© BGR.com © macworld.com

Page 7: 101 Comp. Graphics Recent Advances in CG ONLYhpcg.purdue.edu/bbenes/classes/CGT101/lectures/CGT101-06-Recen… · CGT 101 Recent Advances in CG Bedrich Benes, Ph.D. Purdue University

© Bedrich Benes

Augmented Reality• Challenges

• 3D positioning precisiondepends on sensors and cameras

• Speed of renderingdepends on the CPU/GPU

• Data availabledepends on the internet connection

© Bedrich Benes

Non‐photorealistic rendering

© Bedrich Benes

Non‐photorealistic Rendering• Usually we aim for photorealism• Always we aim for fast displaying

• In some contexts non‐photorealism is better

• Stylized rendering• Artistic rendering

© Bedrich Benes

NPR• Toon shading

displays objects similar to cartoon animation(strong color changes, silhouettes)

Page 8: 101 Comp. Graphics Recent Advances in CG ONLYhpcg.purdue.edu/bbenes/classes/CGT101/lectures/CGT101-06-Recen… · CGT 101 Recent Advances in CG Bedrich Benes, Ph.D. Purdue University

© Bedrich Benes

NPR• Artistic rendering

attempts to simulate artistic style

© Bedrich Benes

NPR• Sketch‐based rendering

simulates human sketches

Interactive Line Art Rendering of Freeform Surfaces. Computer Graphics forum, Vol 18, No 3, pp 1-12, September 1999.

© Bedrich Benes

NPR• Illustrative visualization of volumetric data

• Needs smart processing of volumesand feature extraction

© Anna Vilanova© Bedrich Benes

NPR• Nowadays

Strongly supported by deep learning

• Many of these approaches are image processing techniquesSupported in image processing softwaree.g., Adobe Photoshop

Page 9: 101 Comp. Graphics Recent Advances in CG ONLYhpcg.purdue.edu/bbenes/classes/CGT101/lectures/CGT101-06-Recen… · CGT 101 Recent Advances in CG Bedrich Benes, Ph.D. Purdue University

© Bedrich Benes

NPR

© Bedrich Benes

Deep Learning

© Bedrich Benes

Deep Learning• DL belongs to machine learning (ML)

and ML belongs to artificial intelligence (AI)

• Game changing technology• DL

• Works on large datasets (usually)• Requires huge computational power (usually)• Provides end‐to‐end solutions

© Bedrich Benes

Deep Learning• Traditional ML creates rules for decisions

e.g., if is it round, it is a circle

• DL uses massive datasets to find answers• Show it 10,000 cats and it will recognize cat

Page 10: 101 Comp. Graphics Recent Advances in CG ONLYhpcg.purdue.edu/bbenes/classes/CGT101/lectures/CGT101-06-Recen… · CGT 101 Recent Advances in CG Bedrich Benes, Ph.D. Purdue University

© Bedrich Benes

Deep Learning• In the core is a deep neural network

© M. Mitchell Waldrop © Bedrich Benes

Deep LearningIt is a two step process

1) Training

2) Inference

© Bedrich Benes

Deep Learning1) TrainingRequires a DL (with certain architecture)Requires initial weights of neuronsRequires large dataset of labelled data

© Bedrich Benes

Deep Learning1) Training

zero

one

nine

Training data

© MINST dataset

Labels

Page 11: 101 Comp. Graphics Recent Advances in CG ONLYhpcg.purdue.edu/bbenes/classes/CGT101/lectures/CGT101-06-Recen… · CGT 101 Recent Advances in CG Bedrich Benes, Ph.D. Purdue University

© Bedrich Benes

Deep Learning1) Training

The DL is shown batches of pairs [label, data]DL assigns its weights by using back propagation algorithmFor large datasets this can take daysGPUs are commonly used for this task

Cat

© Bedrich Benes

Deep Learning2) Inference

The DL is shown an unknown data setIt will assign it a label

Inference is very fast (in orders of milliseconds)

Cat

© Bedrich Benes

Deep Learning• Why it all works?• The DNN generalizes the data• Each layer holds higher level of abstraction

• Overfitting problem – it can “memorize” the input data and will not generalize at all(if you memorize the homework you will fail the exam)

© Bedrich Benes

Deep Learning• Convolutional NN

• Image classification• Image segmentation

• Recurrent neural networks • sequences in time• Translators

Page 12: 101 Comp. Graphics Recent Advances in CG ONLYhpcg.purdue.edu/bbenes/classes/CGT101/lectures/CGT101-06-Recen… · CGT 101 Recent Advances in CG Bedrich Benes, Ph.D. Purdue University

© Bedrich Benes

Deep Learning• Generative Adversarial Networks

• Can complete images

© Phillip Isola et al.

https://www.youtube.com/watch?v=5w685udM838&feature=youtu.be© Bedrich Benes

Deep Learning ‐ Reinforcement Learning

© www.openAI.comhttps://www.youtube.com/watch?v=kopoLzvh5jY

© Bedrich Benes

Food for thought… 

© https://autonomousweapons.org/https://www.youtube.com/watch?v=TlO2gcs1YvM © Bedrich Benes

Use what you have learned to make the world a better place