Confessions of a researchaholic

August 31, 2018

Practicing perspective drawing

Filed under: Real — liyiwei @ 3:06 pm

We went to the intersection of Park and Almaden to draw the street palm trees.
To help provide guidance, Daichi instructed me to frame the picture first followed by high-level structures: vanishing points/lines, horizon line, and placeholder of the tree trunks.
This is a relative easy one for me, once I refrained from incorporating too many details. After finishing the drawing, I noticed the background buildings spontaneously form up.

We sat on the benches in front of the movie posters of the center of performing arts.
A homeless person loitered around speaking loudly to himself without bothering us.

August 26, 2018

Git research source

Filed under: Real — liyiwei @ 3:53 pm
Tags: ,

Below is a summary of my suggestions of using git to manage research source, based on numerous discussions I have had with my collaborators across multiple projects at multiple institutions.

Always revision control your code. There are multiple options, but at this moment of writing, git beats alternatives such as svn and perforce.

Use github for public repos and bitbucket for private repos.

Start with existing code if feasible. If the code is already under git, use submodule as a component or fork/branch as an extension. Otherwise, convert the code into git, and preserve revision history whenever possible (e.g. github/bitbucket can help you convert svn repos into git repos).

Use multiple branches for different versions of the code, such as stable branch for release or a personal branch for experiments.
Use multiple remotes across organizations and time-frames.
For example, during your internship with a company, use an internal corporate github repo, and sync it with an external bitbucket (private) repo at the end of your internship so that you can continue the project at your school. When you are ready to publish your code, mirror it to a public github repo.
Keep multiple remotes under the same repo for easy management.

For paper drafts, you can create an overleaf git branch for more WYSIWYG-style editing while retaining all the benefits of branching and revision control.

Keep one (or at most a few) sentence(s) at one line, to avoid false alarms from spell checkers (broken sentences) and excessive (line-based) diff by git.
Note that Latex treats line breaks as spaces.

Chongyang Ma (@ Snap Research) has recommendations for industry research code practice.

August 20, 2018

Shadow QA

Filed under: Real — liyiwei @ 2:41 pm

During the QA after each paper presentation, it is common for the advisers to help the students, especially first-timers, to answer (or translate) questions.

I used to sit on the stage, which I dislike as the bright podium light can shine directly into my eyes. I once shielded my eyes with a hoodie, and people commented that I looked like a gangster.

Then I switched to sit on the front row and step up to the podium during QA. People can see my face; I prefer to remain invisible.

Finally, this year one of my papers was presented in a large, dark ballroom.
A narrow light beam casted over the speaker and the podium was right on the edge of the stage, which allowed me to sneak aside without being seen.
This is my favorite stage design so far.

August 16, 2018

The coolest I have seen in SIGGRAPH 2018

Filed under: Real — liyiwei @ 7:39 pm

Coolest demo in real-time live: AI + animation in addition to sketch-based modeling in Teddy

Very cool VR format for anime/manga by Square Enix

Coolest exhibition: the looking glass autostereoscopic display, the game demo, for the first time, made me think that consumers might finally want to buy a 3D display at home.

August 11, 2018

Drawing a fireplace

Filed under: Real — liyiwei @ 7:47 pm

I found the perspective and framing very challenging.

Drawing a playground

Filed under: Real — liyiwei @ 9:27 am

I was thinking about targeting the abstract sculptures on Park Ave, but Daichi said such structures are not suitable for practice as they can be drawn in anyway and still look reasonable.

So we went to the park and drew the small playground inside.

My high level perspective was incorrect from the beginning, so Daichi asked me to restart from scratch.
He was worried about the homeless people roaming around. After he came back to work, I continued to finish the drawing to a reasonable degree. Two homeless (looking) people passed by and friendly asked me if they can see what I was drawing.

I promised Daichi to draw in Vancouver next week.

August 10, 2018

成功地將作者漢字姓名塞入一篇 SIGGRAPH 論文

Filed under: Real — liyiwei @ 3:47 pm

After years waiting for the right combination of paper format and authors, I finally managed to insert 漢字 author names to a research paper which now resides in the ACM digital library.

August 5, 2018

OKR 2018

Filed under: Real — liyiwei @ 5:03 pm

For OKR 2017, I rated myself 50 percent:

  • Published 5 top papers (with 2 student first authors)

  • Shipped no product, but has at least one in the pipeline (an advantage of working for a company instead of a university is stronger/mandatory incentive to ship)

  • Helped my second HKU PhD student publish her first SIGGRAPH paper so that she can be on track for successful graduation

  • Started regular sketching exercises by finding a mentor

  • Read several books, two are very good (Chaos Monkeys and Principles), but have yet read the Book of Change and the Baroque Cycles.

The default period is from July 2018 to June 2019 unless stated otherwise.

Publish top papers

Publish at least 4 papers in top graphics/HCI/vision/ML venues: ToG (including SIGGRAPH and SIGGRAPH Asia), CHI/UIST, NIPS/ICML, CVPR/ICCV.

Build top products

Ship at least one product as a manager, one as a developer, and file at least two patents.

Develop top talents

Help at least 4 students publish in top graphics/HCI/vision/ML venues individually.

Be a better manager by helping the team not only ship at least one product (above) but also publish at least 4 papers in top venues with at least 4 student first authors (without double counting with my own above).

Pay off the remaining TA unit of my second HKU PhD student; sorry, I should have stay put a bit longer.

Have fun, be awesome

Practice sketching to a professional level.

Pick up a Middle Eastern language to elementary school level, complementing the one Asian and one European language I already know.

Practice meditation to the level that I can enter the God mode immediately, anywhere and anytime.

Finish the book of change, to renew my (now rusty) Chinese and crack the only Chinese classics that my grandfather did not during his life time.

Finish the baroque cycle, which I lost focus in the middle of the first volume (quicksilver).

Reduce my steady heart rate to the previous level – less than 40 beats per minute, and continue improving the art I have started learning for about 2 years to the stage of practical applications.

August 3, 2018

Tree drawing

Filed under: Real — liyiwei @ 12:30 pm

The proportion is not right as I did not plan enough.

To draw tree leaves, I can squint my eyes to low pass filter and use hatches to convey both the values/tones and textures.

I should draw only the tree, not the background, to reduce visual clutter.

Theme: Rubric. Get a free blog at