So, without further ado, let's get on with it!
About the Conference
Siggraph is a top-tier international computer graphics conference with a long and storied history. It is held annually in North America, and attracts a large number of attendees from academia, industry, and fans of computer graphics/animation. Siggraph Asia is a second annual conference (also run by ACM Siggraph) in its 8th year, and is held different Asian cities each year (Shenzhen last year, Macao next year). According to the organisers, there were over 7000 attendees, from over 50 countries present.
This was the first time I'd attended an international conference, as well as being my first Siggraph/graphics conference too. While I've been following the news from Siggraph (via the BF's activities there each year), it was still very much a leap into the unknown, with very little to go on about the general standard of presentations, the way that things worked/ran at the conference. Nevertheless, it was exciting to be a part of the conference and presenting some work too at the same time!
General Impressions & Highlights
* I was surprised by the volume (or rather, the lack) of comments/discussion following talks
Basically, it was much lower than I'd expected. Like 1-2 comments max in most sessions (some of the Pixar talks basically had none; perhaps because people in general felt scared to have to step up to the mic in front of so many watching eyes? For the Good Dino presentation, I'll admit that I was actually within moments of actually going for it, but with all eyes roaming like that, it's admittedly not easy!)
So, relatively speaking, in retrospect, my Grease Pencil presentation actually got quite a good response. Sure, most of the Q's during the session were by the session chair, with one by an art student at SCAD or something like that IIRC. A few other people came up to speak with me about GP afterwards at the conf, including randomly during the reception, which was kindof surprising at the time, as I'd thought that everyone who would had already had their say by then :)
* Speaking of the reception, this happened...
Sessions Attended + Favourite Sessions
I remember the night I first saw the schedules announced on the conference website back in September, and going: Aaaaggghhhh!!!! :D :D :D :D :D Soo much cool stuff that I want to see, but how do I fit it all in?! It quickly became apparent though that it was going to be impossible to attend absolutely every session I was interested in, as especially on the first two days, there were quite a lot of 'em, with quite a few of them occurring at the same time!
As a result, I ended up trying to whittle down a list of "key" things that I most wanted to see - some of these decisions were driven/forced by purely pragmatic/scheduling reasons (i.e. key factors were conflicts with other sessions I was more interested in, and conflicts with my presentation/occurring so close to it that I'd practically end up having to give it a miss, since truth be told, I usually ended up needing to be MIA for 1-3 hrs before a giving a talk/performance... #stagefright ;)
But, in the end, despite the best of plans, sometimes life is messier, and in the process of improvising, you end up stumbling on gems that you may have otherwise missed :)
So, what sessions *did* I end up going to:
1) Opening Session for Symposium of High Performance Computing and Visualisation
2) Animation - Technical Briefs Session
3) Keynote - Ronnie del Carmen
4) Colour and Sketching - Technical Papers Session
5) Sanjay's Super Team - Production Session
-) Tradeshow, Art Gallery, Emerging Tech, Posters
6) Renderman Art and Science Fair
7) Electronic Theatre Screening
8) Good Dinosaur - Production Session
Of these sessions, my favourites were (in descending order):
4) Colour and Sketching papers
I initially attended this session for the Sketching papers only (in particular, for the Autocomplete Sketches paper that was getting a lot of buzz). It was interesting listening to how the researchers described their work (interesting to note here that both projects were from students based at HK universities!), and that both described techniques of relevance to Grease Pencil. That said, on reading the papers, it turns out that getting them into a functional state in Blender would still require quite a bit of work for various reasons (as outlined below).
Room 301 - Technical Papers Room. (Source: Official Conference Photos)
- The "Closure Aware Sketch Simplifying" (i.e. doing automated inking/cleanup of sketchy lines to a clean line) is a reasonably good quality method but is painfully slow (we're talking several minutes here... a fact only really mentioned in the paper). Nevertheless, it gave me some ideas of things to try for Grease Pencil to make inking work nicer for those who do it (though personally, I'm really not a fan of cleaned up linework anymore... too sterile, bland, and it makes the art look like it was just churned out fully-formed by a machine)
- The "Autocomplete" paper as we all know is quite impressive, and the high level ideas behind how it works are actually quite "simple" to get your head around. However, having been doing some close reading of the paper, it sounds like there's quite a lot of work required to implement yourself one of these; in particular, it's a case of "the devil's in the details", with lots of tricky points in the matching/classifying algorithms in particular. But, the promising part is that, as confirmed by the authors during question time, since this technique is purely just vector-based (i.e. it just operates on sets of point data), it's perfectly possible to extend it to 3D. Just what we want to hear :) At least one thing from the paper will be making it into Grease Pencil as a first step - I'll be looking into adding a processing step where the strokes get resampled to have be more regularly sampled.
I was pleasantly surprised by the first two colour-related papers (*). This was because they ended up talking about some of the topics I've been tangentially encountering during my PhD work on highlighting techniques (e.g. perceptual colour spaces - and how to convert/show things in them, and how well we can detect colour contrasts). It was quite exciting for me hearing how these fellow grad students went about attacking these problems; in particular, their general approaches with regard to what experiments they ran (and how) are a lot more compatible with my own intuitions and favoured approach (for the good of the industry, to actually solve the damned problems once and for all!) vs how my supervisor seems to want to attack things (i.e. heavy upfront planning, storytelling, . It was also pretty entertaining hearing how they hacked monitors to do what they needed: the perceptual colour space guys took apart the projector to expose its 6-colour LED array, which they proceeded to hook up some wires to gain control over it (apparently, their butchered apparatus was difficult to transport. so we couldn't get a demo); while the contrast guys stripped the back of the monitor off so that they could get rid of the backlight, and install in its place an uber-powered lightsource that could go up to 5000 Lux IIRC! Oh yeah! Evil hardware hacking!
(*) The third colour-related paper was a bit more of a novelty, and from the samples provided after the talk, the technique still has a long way to go. The idea was to develop an algorithm which could print images on an ansiotropic metal surface so that you see two different versions/images depending on how you've got the metal rotated. For example, you might have a coloured umbrella in one orientation, and a greyscale one in another. Unfortunately, from the samples I got hold of, it really doesn't seem to work/hold up that well. It really depends a LOT on the type of lighting you've got. With most of the lights I've tried it under, the effect is really not that evident at all :/
8) Good Dinosaur session
Of the Pixar sessions I attended, this one by far had the most interesting and juicy tech discussions. Had a nice chat afterwards with the sets supervisor (who was one of the presenters) about their tools/pipeline/and tech specs (rendertime/memory/etc.). I scribbled down quite a few notes about how their set dressing tools/pipeline and cloud generation kit worked, which I'll put up sometime.
1) Visualisation Keynote by Profession Kwan Liu Ma (UC Davis)
This was a session I kindof "stumbled" into, and am glad to have made it to. It was a nice intro into the world of scientific visualisation and the kind of work that's going on in that space. Some key highlights from the work his group has been doing IIRC included:
- Creating a system for automatically figuring out what the optimal lighting/rendering setups for a dataset would be to make the visualisations more useful/legible,
- Demos of how simple Ambient Occlusion (IIRC) effects can be used to improve the quality of renderings of scientific datasets (i.e. you could see a lot more detail and structure in the supernova datasets he showed... those were really pretty and interesting things, especially when they are in motion!)
- A discussion about the concept of doing lightweight "diagnostic" visualisations using all the raw sim data (before it all got chucked out) to generate a set of images that researchers can then investigate interactively by using sliders. Unfortunately, I didn't entirely follow the arguments here about the rationale for this or how it all worked; but, in any case, visualisation is a powerful tool that you shouldn't overlook when trying to debug stuff, and the more interactive you can make it, the better!
Sessions I had intended to attend, but didn't end up going to for various reasons:
* Image and Video Processing Tech Briefs
Registration took much longer than anticipated in my "worst estimate" case, and in a piece of pig-headed disorientation (* 1) failed to register that the entrance to room 501 was around the corner from where the signs/maps indicated. I originally wanted to attend this to gauge what the standard of presentations and order of proceedings were for the tech briefs sessions, as well as listening to a few talks which had some ideas of some minor relevance for GPencil work. Oh well, there are still the papers I can look into. (On the other hand, missing this session meant that I was suddenly able to attend a very interesting one instead, and one that I'd originally had to pencil out due to scheduling ;)
* Tech Papers - Fast Forward
Truth be told, I was tired from the day's activities by that point, and wasn't sure where I'd manage to find something to eat in < 1 hr, to be able to make it to the event. So, while it would've been good to see how people do these rapid-fire presentations (AFAIK, you've got < 1 minute to attract people to your talk), I didn't end up going.
* Abstract Thought Production Session
It was a combination of - if I keep that morning free, I'll at least get *some* brief sightseeing done (vs having visited Kobe, but not really having seen even the basic tourist-y destinations to get a feel for the place) + yeah, I suspect the challenges were more art/style based than technical wizardry, so I might give it a miss
* Big Hero 6 Production Session
The title and abstract sounded intriguing enough - would it be about the CG side of the tech stuff, or more about the robotics they based Baymax on? However, since this was within 2 hours of my talk... yeah... the "serious" business comes first I guess...
* One of the follow-up HPC+Viz sessions
After enjoying the keynote, I wanted to try and see some of the sessions where researchers presented their visualisation projects (weird + pretty pictures, yay!) and/or some of the techniques used to do so. However, I couldn't get the session times to line up with the free gaps in my schedule... It's a pity, as this was some interesting stuff that was a new segment at this year's conf :(
* Many of the short film/exhibitor talks held over on the "Exhibition Hall" side
Yes, I probably shouldn't be quite so lazy, but somehow, the Exhibition Hall seemed quite "far away" and "out of the way" from everything else. Also, the times of those things were always inconvenient/clashing with other stuff.
(*1) In my defence, I'd just arrived in the building for the first time, having just had brisk walk down the winding walkway between the exhibition hall and the conf centre and climbing several flights of stairs, all while wearing a backpack weighing somewhere close to 10kg's from the combination of Laptop (~3-4kg) + Tablet (1kg) + Camera (2kg) + Adapters+Cables (1-2kg) + Water Bottle (1kg). Note to self - 1) Do NOT store camera in the same bag as a laptop ever again; 2) Search for a lighter laptop (but not too much smaller) with smaller adapters
Sanjay's Super Team
I ended up seeing this short 3 times at the conference - twice at the presentation about it, and again the next day at the Renderman talk. It will be premiering to general audiences tomorrow (hence my rush to try and get this post up today... I'll be filling out the rest of this post later) in front of The Good Dinosaur.
General Impressions (some spoilers present):
* I love the whole East-West juxtaposition thing they've got going here. "Worshipping the tv" = genius!
They had an interesting discussion about "camera language" (e.g. telephoto + claustrophoic + closed + square + plain for the living room, vs wide angle + vast + round + dramatic for the Asian-inspired temple scenes). It's always interesting hearing about these sorts of subtle things and how they're used for
(Spoiler) Especially cool is how they had the two of these looking almost exactly like each other - down to the antenna = candles thing on top
* It's also really touching how much of a personal this film is (watch out for the credits sequence - I'll admit, I did tear up a little when the photos of the director and his dad came up, with the lovely soundtrack in the background at that point). I also particularly loved the little bit of acting/expressions of the father towards the end of the film - that look of resignation ("Sigh... I have failed...") was really well done.
* Thinking about this film a bit has made me realise how I too can relate to some of the core themes of this short - that of the conflicts felt by kids growing up in households with "immigrant" parents who have a different cultural background that of the country where you're living. Especially of the sense of resentment about how they're doing "those things" again (in my case, Asia Dynamic/Downunder on Saturday morning, and local Chinese radio from 8-9pm on Sundays).
* The use of sound, particularly in the epic dream sequence is really awesome, and IMO is the source of a great deal of its power. It's also (spoiler) a bit of a major plot point...
* There are a lot of things that you only really start noticing and appreciating on subsequent viewings. Especially after hearing about how they did what they did, and how they arrived at the creative decision used about the (spoiler) "cosmic temple transition". Certainly, I didn't notice most of the things they described as happening when I first saw the film, but did notice on subsequent viewings.
(Spoiler) On first viewing, I only noticed that the transition moment suddenly went to this really awesome looking "white dots zooming" effect; only after the talk did I realise the extent of what was happening (i.e. the dimensions of the set were changing to give the scene the necessary "oomph" and grandness/epicness of "space", while the way it was rendered was simultaneously changed for artistic reasons).
... (TO BE CONTINUED) ...