partiallydisassembled.net

Unidentifiable objects

2007-12-07 17:45:29

I recently had the good fortune of winning a (Google) swag of Google- and RMIT-related merchandise for doing a short presentation at uni. This bag was chockas with crazy things like RMIT golf tees, an antigravity wine holder made of rare tree, a tie featuring the Victorian coat of arms, an RMIT silk scarf, and all the Google pens, notebooks, post-its, baggies, and etc that seem to now litter every conference showbag. I now call on the assistance of the Internet to identify the purpose of this Google keyring. It's fairly large, has a solar panel, lots of circuitry inside (about 10 cm^2 surface area), and occasionally blinks the Google Talk logo. It doesn't blink regularly, glow in the dark, or respond to light/dark, clapping, voice input, or brainwaves. What does it do? Should I leave it unattended in my house?

Simon Wittber writes:

Are you able to post some pictures?

rachel writes:

Have you tried twirling it around? Reprogramming it? (a la Defcon 15) Sitting on it? Using it to jimmy a door? Scraping glue off failed craft projects? (can i have the tie?)

Simon Wittber writes:

Just had an idea... does it respond to 2.4Ghz radio (eg wifi)?

Alex writes:

I liked the wifi idea, but alas, neither laptop nor microwave seem to excite it any more or less than than does the random patterns of the ether in my house. I can kind of read half a model number on the inside, trial-and-error googling turned out that it was just the model of solar panel: http://www.5877.com/com/sunshine%2Dsolar/cp_detail.php?id=70075&nowmenuid=508 62&cpath=17296:18254:&catid=17296 (unless some Chinese-speaker can confirm that the text details the additional psychological effects induced by this model solar panel).

cobungra writes:

It's too late. By now Google have indexed your kitchen.

Beowulf

2007-11-29 23:38:34

The first, most surprising aspect of this film is that it's all animated. OK, so it's not like I'd been tracking every review of it before going to this afternoon's IMAX 3D session, but I had watched the trailer, the Margaret & David movie show and read about it in Wired magazine. None of these sources mentioned that the whole thing -- including actors -- is CGI (maybe they thought we wouldn't notice?). Mostly I was going because the trailer was so incomprehensible and I wanted to see what it was all about, and also because Wired said that the 3D technology is all new and not sickness-inducing these days. First, the 3D: from what I understand, it's a film projection from dual lenses that are polarised at 90 degrees to each other; the audience wears cheap plastic polarised glasses that separate the image. By far the best use of 3D was in the studio logos before the film, such as the Paramount mountain. After the studio logos, the film begins in a tavern, and everyone's making merry. Well, there's a diorama of everyone making merry. It reminds me of the opening sequence of Team America, in which the camera pulls back from a silly marionette parody of France into a slightly-less-silly marionette parody of France. Except that in Beowulf, the camera never pulls back; the whole film looks like a diorama. Middle-to-long shots are clearly separated into separate "depth layers", and each layer looks absolutely flat. Especially the people. It looks completely absurd. Moreso due to the appalling animation. Clearly this is the "we motion-captured it and then fixed it" school of animation, which looks worse than the animation in current computer games. It's been a long time since I saw Final Fantasy (the last animated movie trying not to be?), but this was no improvement in the animation department. There are some big-name actors in this film, and I get the feeling that at some time some of them gave great performances. These performances were then munged into the computer and spat out backward and wrong. In imaging, the rendering was at times impressive. Especially in the close-ups of the male characters and Angelina Jolie. The Beowulf character himself had all kinds of facial hair and imperfections we've not seen in CGI before. The female leads did not get nearly as much attention, and at best resembled plastic dolls, at worst resembled computer-generated plastic dolls (the one exception was Angelina, who was rendered with as much care as the lead men). The non-lead characters had clearly less rendering and modeling time afforded to them, and as such rarely got a close-up. Unfortunately, this relegated them to the two-dimensional flat-land of the mid- and long-shots. No, they still haven't gotten the eyes right -- of anyone. Everyone had dead eyes. Oh, the story's ok. Easy to follow, unlike the poem, I'm lead to believe. Extremely predictable. The monsters look good, but then, CGI artists have had "slimy" nailed since Gollum. Browsing through the IMDB user reviews, it seems most people agree with me that this would have made a pretty good live-action or live-action-on-CGI-backdrop movie. The latter can really, really work -- look at Sky Captain and the World of Tomorrow; that film is amazing. I think most of the 3D problems stem from the limited depth available (the whole scene must be scaled to some "real-world-cinema" depth) which truncates the depth information of relatively thin objects like people. Most scenes were completely in focus as well, which probably doesn't help the illusion of depth at all. Combine a large depth-of-field foreground with an out of focus background and you get people talking in front of a painting of a landscape, instead of a conversation on the foothills. Maybe the 2D version, without the many 3D artifacts would be more acceptable, but I'm pretty sure the animation would completely ruin the whole thing anyway. Verdict: enormous waste of a good cast, huge FX budget and a decent script. Avoid unless you have a fetish for talking dioramas.

VBOs, VAs, DLs... to interleave or no?

2007-11-15 00:40:25

Anyone watching the pyglet-commits list will have seen my recent meddling with buffer objects. I have a fairly nice abstraction to VBOs/VAs and interleaved/serialised data now, so swapping one for another is straightforward, and all the ctypes messiness is taken care of. What's fastest? I ran a simple test on a fixed number of triangles (250,000) in which each vertex is shared by 4 triangles, and triangles are always drawn from an indexed array. The results -- click for Spreadsheet -- are surprising, at least to me: * Display lists perform the worst, no matter what. * Interleaving a VBO gives a *massive* improvement if the indices are also in a VBO. * There is no benefit in keeping the indices in a VBO if the vertices are not also in a VBO, and very little benefit even if the vertices are in a VBO. * For this test, the difference between the best performer and the worst, even discounting the terrible display lists, is the difference between 142 FPS and 31 FPS. It really is worth spending time fudging around with different data layouts, apparently. This was run on a 6600GT on Windows. My friend with a 7800GTX noticed no difference between interleaving or not. If you want to play, run experimental/buffer/torus.py from r1416. The output is, er, *space*-separated CSV suitable for spreadsheeting, but not really for human readability. Sorry. I'm sure you can figure it out.

Richard Jones writes:

The display list numbers are a constant 30 FPS for me - CPU bound, I guess. The best result for my 7800GT (by over twice as much as the second best) is "True True static True static" for 256 FPS. The actual results without display lists: interleaved vbo usage index_vbo index_usage fps False False --- False --- 33 True False --- False --- 34 False False --- True static 33 True False --- True static 34 False True static False --- 85 True True static False --- 105 False True static True static 85 True True static True static 256 False False --- False --- 30 True False --- False --- 30 False False --- True static 30 True False --- True static 30 False True static False --- 30 True True static False --- 30 False True static True static 30 True True static True static 30

Richard Jones writes:

Sorry, pasted the less-cleaned-up version there :(

Alex writes:

More traditional results obtained on my iBook (ATI RageM7). VBOs are not supported. Display lists were corrupted for the above number of triangles, and everything ran at 1-2 FPS. I lowered the vertex count to 10,000 (approx 20,000 triangles, compared to 1,000,000 triangles used above). DL Interleaved FPS False False 50.8 False True 51.0 True False 73.18 True True 73.18 Display list is ideal, and interleaving makes no significant difference.

Tray icons

2007-11-13 14:44:16

Let's (because we're procrastinating) enumerate the Windows tray icons I've got at the moment: * iTunes * Pidgin * Visual C# * Steam * Network (a disconnected network card) * Volume * Remove hardware (nothing plugged in) * Another volume control (courtesy Creative) * nvidia settings Some of these are obviously redundant, and an argument could be made for putting nvidia settings in, I don't know, the control panel? That leaves: * iTunes * Pidgin * Visual C# * Steam * Volume Besides the volume control, these are all applications that _also_ appear in the task bar. WTF are they doing in the tray? Obviously the Mac OS X dock is so awesome (no it's not) that everyone's now emulating (poorly) on Windows. But they left the (actually awesome) task bar there. Awesomeness.

Austin Wood writes:

I've had issues with useless tray icons for years. Machines fresh from an OEM are the worst. (I'm looking at you Dell -_-)

TEN HD is not full HD

2007-11-04 21:20:07

It bugs me that there are _so many_ lies and misleading statements in the advertising for the upcoming TEN HD channel (channel 1, switching on in December). Most of the ad content is duplicated in this press release, if you're not the sort to watch commercial TV. First, the repeated use of the term "full HD" to describe their 1080i signal. Since 1080p televisions have appeared in Australia, the convention has been to describe 768p/1080i as "HD" and 1080p as "full HD". TEN is misleading any customer who shells out the extra money for a "full HD" TV after seeing one of the TEN HD ads, when an "ordinary" HD TV will suffice for the signal they're putting out. From the press release: "TEN is the only Australian network transmitting the globally-recognised pinnacle HD broadcast standard: 1920 pixels by 1080 lines interlaced (1080i) and 5.1 Dolby Surround Sound.". As I write this (9 PM on a Sunday night), all three of the commercial stations 7, 9 and 10 are broadcasting in 1080i in Dolby surround. "Bringing the cinema into the lounge room; movies have never looked better at home". Yes they have. Anyone with a 1080p video player, such as a Blu-ray or HD-DVD drive, or a PS3, or a home computer has seen a better-looking movie at home. "TEN remains the only Australian network broadcasting major live sport in HD". Even though I never watch sport, I can see this is pox. SBS, ABC2 and 7 regularly broadcast HD sport. "The ABC broadcasts in 720 progressive (720p), which is superior to 576 but of lower resolution than 1080i." That 1080i is higher-quality than 720p is debatable (in fact, it often is debated; it's just not a very interesting debate). "The SBS transmits in 576p." This is untrue, movies and the news are broadcast in 720p. "With TEN-HD, TEN becomes Australia's first television broadcaster to break away its HD signal [from the simulcast transmissions alongside SD and analogue]". No, ABC2 and SBS News beat them to it by several years, both providing alternative and, in ABC2's case, timeshifted content from ABC and SBS. ABC-HD also occasionally broadcasts different or timeshifted shows than ABC, not counting the weather and music broadcasts, though these do not appear in any TV guide. TEN: the only way you can improve your signal above the other networks is to reduce the number of in-program ads, eliminate the TEN logo during movies and serials, run the _entire_ content of a show without removing subplots to fit in more advertising time, and broadcast TV shows within 24 hours of their US screening if you are advertising that you are (not two weeks later).
login