The Technical Papers section of SIGGRAPH Asia 2018 in Tokyo is shaping up, as always, to be a key part of the conference. But how do authors get their tech papers into a SIGGRAPH or SIGGRAPH Asia conference? And what happens once they do?
To find out, vfxblog asked Hao Li, who is a co-author on two papers accepted at SIGGRAPH Asia this year, how it all works.
Regular vfxblog readers will certainly have heard of Li and his research into digital humans. He is CEO & Co-Founder of Pinscreen, Inc. (which is developing ‘instant’ 3D avatars), Assistant Professor of Computer Science at University of Southern California, and Director of the Vision and Graphics Lab at USC Institute for Creative Technologies.
More about Pinscreen later, but first, the technical papers. This year, Hao is a co-author on two papers accepted to SIGGRAPH Asia:
1. PAGAN: REAL-TIME AVATARS USING DYNAMIC TEXTURES
Koki Nagano, Jaewoo Seo, Jun Xing, Lingyu Wei, Zimo Li, Shunsuke Saito, Aviral Agarwal, Jens Fursund, Hao Li (some more information and links regarding the paper on Koki Nagano’s website)
2. 3D HAIR SYNTHESIS USING VOLUMETRIC VARIATIONAL AUTOENCODERS
Shunsuke Saito, Liwen Hu, Chongyang Ma, Hikaru Ibayashi, Linjie Luo, Hao Li (information on the paper at Linjie Luo’s website)
These papers are the end results of countless hours (and in fact, years) of research. So where does that process start, in terms of submitting a technical paper?
“The bar for SIGGRAPH and SIGGRAPH Asia technical papers is high and the approach for submitting a paper can be very different depending on the type of projects,” says Li. “They can be theoretical/applied and either solve a known problem or something entirely new.”
What to consider before submitting
Before submitting a SIGGRAPH or SIGGRAPH Asia paper, Li notes that, as a general rule, he considers the following things first:
1. Will the reviewer be impressed/excited by the results – not necessarily high-quality renderings, but will the results have a ‘wow’ effect? What is the first impression?
2. Are the technical contributions and novelties significant enough or is it too incremental?
3. Can I position/differentiate my proposed method with existing papers and show convincing advantages?
4. Is the problem interesting? Am I solving a long standing problem, that couldn’t be solved yet? Is my work achieving the state-of-the-art to a well known problem and making a significant impact? Have I introduced a new field that can inspire more work?
Getting accepted
You can find more on how papers are submitted and reviewed here, but Li of course has some inside knowledge about how to get a paper accepted from several years of working in the field.
He says successful papers usually satisfy those questions above, in that:
1. The reviewers must be impressed by the results.
2. The method is new and there are significant contributions made by the paper.
3. The proposed solution is really different or better than existing ones.
4. The problem is exciting, useful, and/or impactful.
“The reviewers should always be convinced why something cannot be achieved yet with existing solutions, and how/why the presented method can solve it,” says Li. “A comprehensive discussion and clear differentiation with related work is always needed.”
Successful papers, Li adds, are generally very well written with very clear contributions. They are also “presented with polished illustrations, figures, and accompanying videos. The evaluations of the method also need to be very thorough and rigorous.”
Each year, too, there are often industry trends and issues that are timely. Li says this can be “favorable for getting reviewers excited, for example, deep learning, VR/AR, and 3D printing.”
You got accepted – now what?
It’s a lot of work just to be accepted, but there’s more to the Technical Papers section than just the paper itself. Presenting the paper at the conference is a major part of spreading the knowledge and generating discussion. This actually begins in the exciting Technical Papers Fast Forward, where authors have less than a minute to entice conference attendees to come and view their full presentation. The Fast Forward at SIGGRAPH Asia Tokyo takes place on Tuesday 4th December from 6pm to 8pm.
For the full presentation, Li suggests the following flow that has been a basis for him and colleagues for some time:
1. Start with some slides to motivate the audience why they should care?
2. Get straight to what problem we are trying to solve.
3. Explain why it cannot be solved previously while presenting prior work, and why it’s challenging.
4. Either give an overview of the method (top-down) or explain the technique from a simple example (bottom-up).
5. Show insanely cool results!
6. Mention some limitations if any.
7. Discuss what’s next and show some future directions.
“The key,” concludes Li, “is to connect to the audience and speak as if you are explaining the work to a friend or colleague, and not sounding like you are reading from a paper. The audience has to be convinced that you know what you are talking about.
Where it might all lead
Technical papers unveiled at SIGGRAPH and SIGGRAPH Asia are diverse, and often lead to continued research and sometimes even real products. Pinscreen is an example of where Li and his colleague’s initial research into digital humans has been taken further.
The company has released an app that generates digital avatars from a single photograph, with photorealistic hair and clothing options. Pinscreen has also launched a facial tracking SDK along with a demo app.
You can find out more at pinscreen.com, and see Pinscreen’s presentation at SIGGRAPH Asia 2018 Real-Time Live! (Pinscreen Avatars in your Pocket: Mobile paGAN engine and Personalized Gaming) on Friday December 7th at 4pm-6pm. Also, check out fxguide’s in-depth coverage of Pinscreen here.
Good luck in submitting your technical papers in the future, and hope to see you at SIGGRAPH Asia Tokyo!
You can register to attend SIGGRAPH Asia Tokyo 2018 at http://sa2018.siggraph.org/registration.
Post a Comment