COMPSCI 715Advanced Computer Graphics

Using Related Work

Research Questions

Group 1:How can feedback (visual stimuli) to increase motivation in HIIT-based exergaming [use exerbike]

Group 2:How can we use music and synchronised game play to implicitly induce HIIT exercises in exergaming [use rowing machine]

Research Questions

Group 3:How can we improve surface reconstruction for 3D building models using context information (self similarity) and characteristic properties of buildings

Group 4:How can we encourage (long-term) motivation in sedentary users using a location-based mobile game (also how can we include goal setting and exercise feedback)

Today’s Mission

1. How do you find good related work for your project?

2. How does a good related work section look like?

3. Examples

Typical Research Paper Structure1. Introduction: What is the research problem?

Introduce and motivate it. Summarize your contributions. 2. Related Work: What have others done? How is it different?

Cite, summarize other solutions & compare it with your own.3. Design: Your solution. Motivate design (what was done and why?)

Provide enough details so others can implement / replicate it. Software architecture (e.g. class diagram)? User interface (e.g. screen diagram)? Algorithms / formulas?

4. Implementation: How have you implemented your solution? Tools and technologies used? Implementation challenges?

5. Evaluation: Explain the methodology you used for evaluation. Present the results. Discuss them.

6. Conclusion: Summarize contributions. Draw conclusions (e.g. implica-tions in practice, guidelines, lessons learned). Point out future work.

Finding Related Work1.Gather papers that might be relevant

oKeyword search with different words(e.g. Google Scholar, ACM Digital Library, IEEE Xplore)

oSnowball search: follow up the references(cited and citing papers)

2.Filter papers that are actually relevant by reading the abstract3.Read the papers

oFirst just scan over it (figures? sections? methodology?)oWhat is the research problem? What are the contributions?oDo they cite related work that is useful for you?oSee http://blizzard.cs.uwaterloo.ca/keshav/home/Papers/data/07/paper-reading.pdf

Peer-reviewed Publications ●Scientists submit papers to journals or conferences

that are peer-reviewed oShort paper ~5 pages, long paper ~10 pages,

journal article ~20 pages●Other scientists, established experts in the field,

review the papers, free of chargeoIndividual reviewers remain anonymous, but the board of

possible reviewers is knownoThey check for: obvious inconsistencies, dubious statements,

good standards of presentation, sufficient degree of completeness, i.e. disclosure of details, appropriate methodology, novelty of the results, and reproducibility

oThey usually will request changes●Many papers get rejected (usually 50%-90% depending on venue)

Reviews and Systematic Reviews●For many important research problems or

applications hundreds or thousands of papers exist●In that case some researchers might have performed

a (systematic) review or state-of-the-art report of that field●Reviews perform an in-depth analysis of the state-of-the-art●State-of-the-art reports are similar, but usually less complete●A systematic review is a special review which tries to incorporate ALL

papers in a field by performing the search systematically (e.g. using certain key words and inclusion/exclusion criteria

●They try to convey insights not obvious from just reading the reviewed papers, e.g. classifications, evaluation, categorisation, common concepts, comparisons, limitations

●Can be a ideal source to start a literature review●Often highly cited●Examples:

https://jneuroengrehab.biomedcentral.com/articles/10.1186/1743-0003-11-137https://www.cs.auckland.ac.nz/~burkhard/Publications/WSCG2012_NguyenEtAl.pdf

Citing Publications● Cite quality assured publications using

proper citation styleo Peer-reviewed (not just pro forma,

e.g. no fake conferences)o Cited by others in the field (means you

should know about it if you are working seriously)o Well-known conferences & journals (e.g. see rankings)

● If you need to refer to a webpage, can use footnote with URL● Take publication year into account (State-of-the art? Obsolete?)● Industry white papers and reports can be cited, but are not quality

assured, so no strong backup for you (“grey literature”)● Textbooks only for very specific things (refer to them with page or

chapter number)● Not Wikipedia (good for your own overview, but unreliable)● Not advertisement material (except when referencing products,

e.g. used for research or as examples of commercial work. Not research results since not independently reviewed)

Writing about Related Work

It is all about analysing previous researchers’ contributions and use results for your own research:1. Summarize the contributions of the good related works in

one paragraph per publication2. If relevant identify methodologies used and evidence

provided (e.g. user studies)3. Identify assumptions and limitations4. Identify common features and differences5. If possible classify and categorise related work6. Structure the related work section appropriately

Provide contextMotivate your research (what’s new?)1. Identify gaps in previous research2. Identify contradictions in previous research3. Identify faulty methodologies used in previous research (e.g. wrong

assumptions, inappropriate comparisons, user studies with multiple variables)

Build on top of previous research1. Extend previous work (generalise, make more efficient/effective,

compare [=> survey paper] etc.)2. Use previous work to perform requirement analysis (learn from other

people’s insights and mistakes)

Evaluate your research1. Compare to previous work and show in what sense your research is

different / novel / better / more general etc.

Using Related Work

Using Related WorkRelated work can be analysed and evaluated using a suitable evaluation metric – in this case the evaluation metric, analysis, and/or evaluation data set can be a novel contributon … the image below shows an example

ExerciseLearn from good and bad example related work sections1. Read the related work section (note: examples are

shortened)2. Identify the following parts:

a. Sources for the related worksb. Contributions of the related worksc. Grouping / Categorisation / Classification of the

related worksd. Differences to own contributions of the project

3. Discuss:What are the good and bad points? How could it be improved?

Good Related Work Example 1https://www.cs.auckland.ac.nz/~lutteroth/publications/VanDykEtAl2012-GLDebug.pdf

gDEBugger4 was one of the first commercial graphics debuggers to become widely available in 2004. It demonstrated many of the features seen in modern graphics debuggers, such as [...] The contribution of gDEBugger is in its pioneering of graphics debuggers in the commercial space, as well as [...]

GLSLDevil9 [14] is a tool specifically aimed at debugging the shader pipeline of OpenGL applications. GLSLDevil provides novel features in that it automatically instruments OpenGL shader code. The instrumented code then outputs extra information that can be used for debugging.

GQL (graphics query language) was created along with a debugging system by Duca et al. [5]. Similar to GLDebug, it enables tracking and logging the state and calls made by an OpenGL program over the course of execution. However, the historical information is only made available through an SQL-like language (GQL) that users have to learn, and there is no direct support for comparing states and highlighting of state differences.

Good Related Work Example 2https://www.cs.auckland.ac.nz/~burkhard/Publications/SpringerLectureNotes2011_NguyenWuenscheEtAl.pdf

The reconstruction of 3D models from 2D images of objects has long been a major research field in Computer Vision. If the object geometry is relatively smooth and the material homogeneous, then surface geometry can be estimated from illumination gradients. This class of algorithm is called “shape from shading” [27]. However, the technique has too many constraints with respect to the object geometry and the environment in order to be useful in general applications. A different approach is to only consider the silhouette of an object from different view points. This “shape from silhouette” class of algorithms is very fast and stable with regard to object colour, texture and material, but is very limited in the object geometries it can handle [20, 11]. A more stable and general approach with respect to object geometry is “structured lighting”, which projects a stable pattern of light onto a surface and reconstructs geometry from its deformation [7]. When combined with a camera both object geometry and texture can be recovered. This approach is used by the popular Kinect controller [28]. […]More general shape reconstruction algorithms can be developed by analysing features in camera images and video data. […] The work by Martinec and Pajdla [19] is closest to our research in the sense that the authors use arbitrary unconstrained input images. […]

Good Related Work Example 3https://www.cs.auckland.ac.nz/~burkhard/Publications/IEEETVCG2012_BockEtAl.pdf

Interactive high-quality rendering of FE models, incorporating the interior of the elements, is a demanding task. Often, the elements are resampled to lie on a common regular uniform grid [35]. While this enables a direct visualization using standard volume rendering algorithms, it also results in sampling-based errors as well as a loss of the easy access to element boundaries. As a consequence, many researchers have investigated rendering techniques that enable a more direct FE visualization by maintaining the relationship between the world and the material coordinate system. Wuensche introduced an isosurface-based approach for the visualization of biomedical tensor field data [37], which has been applied to strain analysis. Schroeder et at. developed a framework for isosurface rendering by tessellating the basis functions into simpler forms which improves the mesh quality [30]. […]To improve rendering quality ray-casting approaches have been proposed. Garritywas one of the first to introduce ray-casting for rendering of unstructured meshes [8]. […] More recently, Uffinger et al. presented a distributed visualization approach that allows volumetric visualization of models to be defined through high-order polynomials [34]. As in ours their approach is also based on ray-casting but, in contrast, they circumvent computation of the world-to-material space transformation by transforming the FE solution into a Cartesian reference space with barycentric coordinates. […]

Bad Related Work Example 1The various documents I read helped me to get a clear picture of what I need to achieve in the time period of completion of the project. Various links on the Wikipedia website regarding Sculpting, Subdivison surfaces and Collaboration helped me understanding the basics for starting the project [...]The available sculpting tools already have become popular because of their completeness for the features provided which are easy for a user to understand and implement. The tool which we/I took as reference is Sculptris. A better definition than just telling it as a sculpting tool is “Sculptris is an elegant, powerful and yet easy to use 3D sculpting software, allowing the artist in you to simply focus on creating amazing 3D artwork.” 3D Collaborative Sculptor aims to implement all the features similar to the features available in Sculptris. Well, might be because of time constraints of the project, it will implement less features compared to Sculptris. Our main aim is to implement the Collaborative feature in the 3D Collaborative Sculptor extending a better solution for the 3D and Graphics related companies as collaboration speed up the process of development with accuracy.

Bad Related Work Example 2

Some of our prototype’s code base was inherited from the research conducted by Zheng et al. [6] a year prior to ours. The framework proposed by the authors would allow user to draw on model, recognize simple shapes (circles, squares etc.), extrude lines and finally bend plain shapes. In addition to that to be able to provide user with the ability to sketch in 3D space we had a look at some possible techniques and their evaluation [3, 4]. Due to the tool we are trying to develop is being addressed towards non-professional and users with the minimal experience in 3D modeling, Kara et al. [3] provide us with a highly important user feedback and expectations from sketch-based modeling. On the other hand [4] provides seven more professional approaches to sketch based interfaces. It also lays out advantages and disadvantages of each, which set us on what we think is a right way to address sketching part of our prototype. […]

Bad Related Work Example 3Murphy-Hill et. al. [7] worked on developing a code smell detector called Stench Blossom which offers an interactive visualization tool to give programmers a conceptual overview of the smells in source code and helps them in comprehending the sources of the code smells. Neukirchen et. al. [8] worked on developing an anti-pattern detecting system which nds anti-patterns in test suites. Their open-source refactoring and metrics tool named TRex was capable of automatically detecting code smells.Stoianov et. al. [10] used a logic-based approach to detect anti-patterns in source code. Their solution uses an infrastructure named Jtransformer that is an Eclipse plugin which generates logic facts that is a representation of java source code.Emden et. al. [11] have worked on implementing a system that visualizes code smells. They implemented a prototype code smell browser called jCOSMO for the detection and visualization of code smells in java source code.

This Week’s Assignment:Write Related Work Section (5%)

Write a related work section for your project (~1 page double-column)● Individual submission, no group work, worth 5%● More references with concise summaries better than less

references with lengthy summaries● Try to structure your related work section● Again, methodology and contributions are hypothetical:

imagine your project is over and was successful● Be professional: try to imitate well-written introductions● Use LaTeX and BibTex, e.g. https://www.overleaf.com/Submit PDF by Sunday 20th August 10pm to assignment dropbox: https://adb.auckland.ac.nzAll the best :-)