Seminar Advances in Computer Graphics

Changkun Ou | Dennis Dietz | Prof. Andreas Butz

Winter semester 2019/2020

​

​

2

Image from "Red Dead Redemption 2"

Information

Lecturer

​

In charge

​

​

What you need

​

​

What you get

​

​

​

Website

3

Prof. Dr. Andreas Butz

​

Changkun Ou <changkun.ou@ifi.lmu.de>

Dennis Dietz <dennis.dietz@ifi.lmu.de>

​

→ Basic Computer Graphics or Machine Learning knowledge

→ English skills

​

→ 2 SWS / 6 ECTS

→ Experience in (real) scientific writing and research

→ Peer review comments in a SIGGRAPH/SIGCHI conference

​

https://www.medien.ifi.lmu.de/lehre/ws1920/hs/

Contents

What you will do in this seminar

​

→ select / be assigned to a general topic today

​

→ write literature review about papers you read (as Introduction and Related Work section)

​

→ give a 15+10 minutes presentation about what you read

​

→ write a paper (that solves your problem, maximum 10 pages)

​

→ give a 15+10 minutes presentation about your write

​

→ submit to a conference and getting peer reviewed :)

4

Schedule

​

5

Week 1

​

6

​

Write down your research topic:

​

1. Your Overleaf paper link

2. You motivation (in 150 words)

3. List related citations (at least 5 papers)

​

Deliver a document in PDF version contains all above in Uni2Work

Week 2

​

7

​

Write your introduction section

​

1. Rephrase your motivation as a formal introduction of the paper

​

2. Enrich your citations (historical research direction, etc.)

​

→ Submit a compiled overleaf PDF document of the paper in Uni2Work

Week 3

​

8

​

Write related works:

​

1. Give a overview to the most relevant literature, discuss their core ideas, core approach and results.

​

2. Enrich your total reference number up to at least 30.

​

→ Submit a compiled overleaf PDF document of the paper in Uni2Work

Week 4

​

9

​

Prepare a 10~15 minutes slides:

​

1. Discuss your motivation of the paper

​

2. Discuss the core ideas of literatures in your related work section

​

→ Submit a PDF document of your slides in Uni2Work

Slack Channel

We use Slack to discuss everything related to the seminar.

Please sign up and join the discussion, you can ask all your questions there.

you must possess a cip.ifi.lmu.de email address. Please inform yourself here on how to obtain one.

10

You should already familiar with either...

• CG Basics
• 3D Basics: Transformation, pipeline
• Geometry: Primitives, Mesh data structure, Surface parameterization, LOD
• Materials: Texture UV Mapping
• Classic Illumination Models: Phong, Rendering Equation, Ray Tracing
• Animations: Particle System, Physical-based Simulation
• ML/DL Basics
• Neural Networks
• Optimizations & Regularizations
• CNNs & GANs

11

Topic I: Geometric Processing

• Graph Neural Network on different representations, e.g.
• Polygonal based
• Volumetric based
• Point-cloud based
• Multi-view based

12

Boscaini et al. 2015

Topic I: Geometric Processing

• Shape Analysis: e.g. 3D objects segmentation, 3D objects captioning
• Reconstruction: e.g. mesh reconstruction from point-cloud, simplex denoising
• Deformable Objects Generation
• …

​

​

13

Song et al. 2017

Ben-Hamu et al. 2018

ShapeNet [Chang et al. 2016]

Topic I: Geometric Processing

How to apply CNN to 3D objects remaining an open research question.

​

Write a paper that reviews all existing way (last 5 years) of doing CNN in 3D geometric objects, the paper is required to cover operators for different 3D object representations.

​

Starting literatures:

[1] M. M. Bronstein, J. Bruna, Y. LeCun, A. Szlam and P. Vandergheynst, "Geometric Deep Learning: Going beyond Euclidean data," in IEEE Signal Processing Magazine, vol. 34, no. 4, pp. 18-42, July 2017.

[2] Rana Hanocka, Amir Hertz, Noa Fish, Raja Giryes, Shachar Fleishman, and Daniel Cohen-Or. 2019. MeshCNN: a network with an edge. ACM Trans. Graph. 38, 4, Article 90 (July 2019)

[3] Hsueh-Ti Derek Liu, Alec Jacobson, and Maks Ovsjanikov. 2019. Spectral coarsening of geometric operators. ACM Trans. Graph. 38, 4, Article 105 (July 2019), 13 pages.

14

Topic II: Shading Techniques

• Global illumination regression
• Ray tracing denoising
• Neural shader
• ...

15

[Salvi, 2017]

[Kato et al, 2017]

Topic II: Shading Techniques

Although real-time ray tracing is in production, but we are not done with ray tracing, yet.

Write a paper that reviews the state of the art techniques for real-time ray tracing, and identify the current challenges (including hardware and software) for future research.

Starting literatures:

[1] Alexander Keller, Timo Viitanen, Colin Barré-Brisebois, Christoph Schied, and Morgan McGuire. 2019. Are we done with Ray Tracing?. In Proceedings of SIGGRAPH ’19 Courses. ACM, New York, NY, USA

[2] Physically-based Modeling and Rendering of Complex Visual Appearance, SIGGRAPH 2019 Outstanding Doctoral Dissertation Award

[3] O. Nalbach, E. Arabadzhiyska, D. Mehta, H.-P. Seidel, and T. Ritschel. 2017. Deep Shading: Convolutional Neural Networks for Screen Space Shading. Comput. Graph. Forum 36, 4 (July 2017)

16

Topic III: Animations (Simulations)

Animation techniques are highly dependent on the use case. Which is why physic-based simulations are most common for object animation, but through advances in this field, it even gets interesting for character animation.

Write a paper that reviews the state of the art progress in physic-based character simulation, and identify the current challenges for future research.

Starting literatures:

[1] Seunghwan Lee, Kyoungmin Lee, Moonseok Park, and Jehee Lee. Scalable Muscle-actuated Human Simulation and Control. 33rd Conference on Neural Information Processing Systems (NeurIPS 2019), Vancouver, Canada.

[2] Xue Bin Peng, Michael Chang, Grace Zhang, Pieter Abbeel and Sergey Levine. MCP: Learning Composable Hierarchical Control with Multiplicative Compositional Policies. ACM Transactions on Graphics (SIGGRAPH 2019), Volume 37, Article 73

​

17

Topic IV: CG-based Applications (Interactions)

With the advances in computer graphics VR and AR applications can enhance existing methods in order to help improve human performance in sports. Therefore combinations of different training methods need to be structured and evaluated.

Write a paper that reviews training approaches through supportive VR/AR systems, identify general challenges among diverse sports for future research.

Starting literatures:

[1] Mads Møller Jensen1, Majken K. Rasmussen1, Florian ‘Floyd’ Mueller2, Kaj Grønbæ. Keepin’ it Real: Challenges when Designing Sports-Training Games. Conference: Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems, At Seoul, Republic of Korea.

[2] Yuji Sano, Koya Sato, Ryoichiro Shiraishi and Mai Otsuki. Sports Support System: Augmented Ball Game for Filling Gap between Player Skill Levels. ISS '16, November 06-09, 2016, Niagara Falls, ON, Canada.

​

18

Topic X: Your own decision

Decide your own topic. Deliver a detailed proposal in the first week delivery package.

In the proposal you must present your idea detailly. Convince us, then you will be approved if we found you idea is suitable for research.

​

Random literatures:

[1] Oded Stein, Eitan Grinspun, and Keenan Crane. 2018. Developability of triangle meshes. ACM Trans. Graph. 37, 4, Article 77 (July 2018)

[2] Liu, Hsueh-Ti Derek, and Alec Jacobson. "Cubic Stylization." arXiv preprint arXiv:1910.02926 (2019).

[3] Peter Hedman and Johannes Kopf. 2018. Instant 3D photography. ACM Trans. Graph. 37, 4, Article 101 (July 2018)

[4] Ran Yi, Yong-Jin Liu, and Ying He. 2018. Delaunay mesh simplification with differential evolution. ACM Trans. Graph. 37, 6, Article 263 (December 2018)

19

Team Up (10min: Intro yourself and share your expertise)

20

Scientific Publishing

21

Scientific conference & Journals in Graphics/Interactions

• SIGGRAPH
• SIGGRAPH ASIA
• EUROGRAPHICS
• Transactions on Graphics
• SIGCHI
• UIST
• ...

22

Step 1:

Choosing a Good Research Topic

23

Choosing a research topic

The most important decision you will make

Think about the time investment you are about to make

What if it is not an interesting problem to begin with?

The easiest way to get a SIGGRAPH/SIGCHI paper:

Be the person with the brand-new way-out idea! (This is hard)

If you can't: checkout "limitations & future works" in a paper

24

Criteria to consider...

1. How important a problem is this?

If it's incremental advance at best, I am less interested.

​

2. How good an idea do I have for how to solve it?

The more I have a good sense for how to solve it, the better

​

3. How original is the problem?

If it is a well-trodden area, I am less interested.

​

4. How well-defined is the problem?

5. How much would the problem lead to interesting continuing future work?

6. How interested am I in the problem, personally?

25

Know your Literature!

You will get detailed question about the literature you read

26

How to research a topic - Search

Here covers almost every important literatures:

• ACM Digital Library https://dl.acm.com/
• Google Scholar https://scholar.google.de/
• ArXiv https://arxiv.org/list/cs.CG/recent/

​

27

How to research a topic - Read

Read in multiple steps

1. skim over abstract and images → Interesting?

2. read complete → get it

3. read in detail → detailed understanding

​

While reading

→ take notes

→ mark text passages

→ what were they doing? how? why? results?

​

Finally

→ check reference papers

28

How to research a topic - Read

Tools

→ Mendeley, JabRef, Zotero, Paperpile, …

​

Why?

→ search and retrieve

→ labeling ('nice', 'bullshit', ...)

→ notes, citations, ...

29

Step 2:

Doing the Research

30

Problem Refinement Process

Find ways to simplify your problem:

​

- Use synthetic data

- Lower the dimension

- Find ways to run quick tests

​

→ ...optimize the research's "inner loop"

​

​

​

Keep a written list so you can track & prioritize them.

​

​

31

Experiment

Evaluate

Ship

Content Management Process

Begin write the paper early!

​

• Helps you sort out the big ideas from the hacks -- and formalize the intuition behind any hacks that remain.

​

• Gives time to craft the "story" that you want to tell.

​

​

A hard problem is a good thing, but a very difficult thing to find a problem that is just hard enough.

​

The problem will have to change if you are going to make any good progress.

32

Step 3:

Writing It Up

33

Tool to Collaborate

- Learn to use LaTeX

- ACM SIGGRAPH Technical Paper LaTeX template → Link

34

Writing is well important

Helps you clarify your own ideas

Well written paper can be appreciated by your audience and well into the future

​

Writing a research paper is different from other technical writing:

- Not just about careful, clean exposition

- Also about what the problem is

- And how you arrived at the solution

​

Do not write bullshit

Simplicity is complicated

You learn how to write paper while reading good literatures (Imitation)

35

Assets (Figures/Tables/Algorithms...)

Think of the figures (assets) as a storyboard for the whole paper: a separate track that stands on its own.

​

Your readers should be able to read through just the figures (assets) first:

​

- They should all make perfect sense

- They should comprise a summary of the entire paper

​

Try presenting your work as a talk with just pictures before writing the paper, then reuse them for the paper's figures.

36

Taking care of your citations

Sources

→ always reference the primary source

​

Indirect

→ Lorem ipsum dolor sit amet [1]

→ Authors et al [2] found out that [...]

→ put references always before the dot [3].

→ multiple sources: [4, 5, 6]

​

Direct

→ only use direct quotes sparsely, e.g. for definitions

→ use correct English quotation marks for direct quotes: "This is a direct quote" [7]

37

Resources

• David Salesin. How to write a SIGGRAPH paper: a guide to choosing a good research topic, doing the research, and writing it up. In SIGGRAPH ASIA 2016. https://dl.acm.org/citation.cfm?id=2988471
• T. Möller et al. Real Time Rendering (4th Edition). CRC Press, 2018. http://www.realtimerendering.com/
• Goodfellow, Ian, Yoshua Bengio, and Aaron Courville. Deep learning. MIT Press, 2016. https://www.deeplearningbook.org/
• Ke-Sen Huang's Home Page, http://kesen.realtimerendering.com/
• Bibliography https://dl.acm.org/ / https://scholar.google.de/ / https://arxiv.org/list/cs.GR/recent

​

38

Caution

​

You will immediately fail the course if you…

​

• Plagiarism
• Miss more than three deadlines

​

You will get better grades if you…

​

• Give awesome presentations
• Writing an outstanding paper in the end
• Deliver everything without missing any deadlines

​

​

39

Don't forget!

​

40