What is Digg?Warning: The Content in this Article May be Inaccurate
Readers have reported that this story contains information that may not be accurate.168 Comments
- dtfinch, on 10/12/2007, -3/+61I bet the 23mb video is a raw, uncompressed video of total silence and blackness.
- D4RKfantasy, on 10/12/2007, -1/+33Ugh, I smell another KGB and Fractal Compressor in this one.
- DevilsRejection, on 10/12/2007, -4/+32get this technology into the hands of independent press agents or known reliable sources and let's get an opinion.
i wouldn't clean my ass with a print out of a press release from anyone. - ThatsUnpossible, on 10/12/2007, -0/+28No, no, no.
However, the resulting euclid video file is only at a resolution of 2x2 pixels. - inactive, on 10/12/2007, -2/+30I just wrote a new codec that will enable all 3 LOTR movies to be compressed into this:
0110
Feel free to download this HD version before digg gets shut down! - sundancekid503, on 10/12/2007, -2/+23I'm downloading the torrent of this right now... damn..... stuck at 80% waiting for the last 0
- inactive, on 10/12/2007, -1/+19Talking Heads would be an awesome name for a rock band.
Oh, wait... - atdigg, on 10/12/2007, -0/+17Even better, I reduced all the Universe to 101010
- tazamore, on 10/12/2007, -1/+17Seems a legit company with a downloadable demo:
http://www.eucliddiscoveries.com/information.php
Seems their compression is optimized for talking heads, like video conferencing, where only some of the image changes from frame to frame. But I'd bet it has nasty artifacts for other kinds of video.
They are proud of their 15 patents. They will most likely try to license this to video conferencing suppliers. I doubt they will release a freeware version and if they did MPAA would probably try to sue them for further enabling piracy. - mrASSMAN, on 10/12/2007, -1/+16No need for a torrent, just type 0110 into Notepad, and change the file extension to .avi. The computer does the rest!
- Olle, on 10/12/2007, -0/+14Seriously, it must be.
They say "EuclidVision compressed a 23mb 30fps reference video down to 1,519 bytes (15,168:1 compression ratio, 3.56 Kbps bandwidth). MPEG-4 H.264 compressed the same video at similar quality to 8,518 bytes"
They claim the test file is 23mb and mb really means mega _bit_ afaik, but MB, for mega byte, is the normal size measure so who knows that they mean with that measure.
Anyway, wether they mean 23 megabit or 23 megabyte, if Mpeg 4 can get it down to 8,518 bytes then it must be silence and darkness. Or a small single picture that doesn't change. - FlyingAvatar, on 10/12/2007, -0/+13Hahahah, 1519 bytes? Riiight.
And I just invented a new audio compression scheme that can further compress a 5mb mp3 to 516 bytes. With virtually no discernable difference in quality to dead people! - Johnny2085, on 10/12/2007, -0/+13This is obviously an April Fool's Joke for Slashdot. It has to get on Digg ahead of 4/1 though, so the slashdot editors can find it and then post their dupes on the actual day.
- Otto, on 10/12/2007, -0/+12When they say 1519 bytes, it sounds like they mean it. Note that they say that H264 compressed the same video to 8,518 bytes.
However, this is clever marketing speak.. The important bit here is the bitrate. Notice that they says the video was 23MB and 30fps, but they didn't give an actual bitrate or length or any other parameters to determine how many frames we're talking about here... However, they DID give a bitrate of the final videos!
So...
1519 bytes * 3.56 Kbps = 3.33347963 seconds
8518 bytes / 19.96 Kbps = 3.33401177 seconds
Which tells us that this 23MB video was, in fact, only about 3.33 seconds long. Which means that it was probably exactly 100 frames of video.
Their compression seems a bit less impressive considered in that context, and you don't need to assume a black screen. It's just that there's not enough frames to perform a really good test of the ratio over time. - threepio, on 10/12/2007, -3/+15You owe me a new keyboard. And a new carton of milk.
And possibly some Kleenex and a clean shirt. - inactive, on 10/12/2007, -0/+11I can only encode seasons 1, 2 and 3 right now: (it works well in sets of 3)
1101 - craigwblake, on 10/12/2007, -2/+13It's not a typo, ~1500 bytes is what you get applying their advertised compression ratio of 15,168:1 - 24,117,248 / 15,168 = 1,590.0084. Similarly, the MPEG4 compression ratio they stated as 2,705:1 applied to the same initial file size results in the MPEG4 file size they also claimed - 24,117,248 / 2,705 = 8,915.8033. Clearly, they actually did mean to say "bytes" there.
Now, whether or not that is realistic is another question altogether. The important thing to realize is that they explain that their compression ratio only provides an 5.6 times improvement over standard MPEG4 compression, which is not too unbelievable considering the particular subject matter, i.e. a still background with a face on it that really only has moving lips and blinking eyes most of the time.
It is rather amusing that they are so proud of their patents, though. - mrASSMAN, on 10/12/2007, -1/+12WOW! I decompressed your numbers into multiple Blu-Ray discs!
Do you have some binary for the entire 24 season? (in HD please) - dragoth, on 10/12/2007, -1/+11From the article:
"There is no company that is solely dedicated to making such effective use of this methodology besides Euclid Discoveries, and the company has effectively shut the door on future competitors because of its diligent patent filings covering key aspects of this underlying technology."
Bah! Humbug. Anyone remember the old video-over-modems hoax that sprang up a few years ago? - FlyingAvatar, on 10/12/2007, -0/+9Amusingly, that's actually what it says in the article.
- mrASSMAN, on 10/12/2007, -0/+9ok fine, i lied. the movie is twice as large as my 16x16 avatar.
- L0phtpDK, on 10/12/2007, -0/+9Actually, they are right:
Their compressor, aka "Yadda-Yadda-Yadda" reads the objects on the screen and gives you the general idea, or as they call it, the 'Jist of it all'. So you gain the same amount of information, but in a ‘slightly’ compressed form.
For example: Euclid Discoveries compressed a video comprised of a reenactment of Lincoln's Gettysburg Address (which would have been aired on TNT, because only them would do that kind of *****). Using their "Yadda-Yadda-Yadda" engine, they compressed the ENTIRE speech into this:
"Some Joe with a beard and a hat saying something about being four scored... when some guys died... yadda-yadda-yadda... The Union is good."
And there you go; a 15,168 to 1 compression of the Gettysburg Address via the patented "Yadda-Yadda-Yadda" engine.
That should quiet all you naysayers. - JohntB, on 10/12/2007, -1/+9millibits? That explains how the resulting file was so small.
- Toallpointswest, on 10/12/2007, -1/+8I'll believe it when I see it.
- coding, on 10/12/2007, -1/+8Yes, and using most compression software you can zip up files larger than 80 gigabytes down to 300 bytes if the file contains just the letter 'a'.
- iloveredcups, on 10/12/2007, -4/+11Death penalty? Ha! There is a typo in your comment. "Mathematical errors and typographical mistakes are offenses that cannot be forgiven (,) and should be considered for the death penalty."
"Should be considered for the death penalty" is not an independent clause. - mrops, on 10/12/2007, -0/+7Actually I let KGB run for 24 hrs since the story was posted yesterday. I didn't get those dramatic results of 460MB to 1.4MB, but I did compress about 50MB file to about 10MB in 24 hours. Best WinZip and WinRar did was 26 and 27 MB respectively. Though both rar and zip saved me 24 hrs, give or take a few minutes, compared to say KGB.
- Knoton, on 10/12/2007, -1/+7Actually that avatar is just under half that at approximately 661 bytes
/nitpicking - inactive, on 10/12/2007, -0/+6"Even better, I reduced all the Universe to 101010"
hmmm that's weird, when i compress life/universe/everything i get: 101111 - jaypee68, on 10/12/2007, -0/+6I'm trying the demo right now. I'm encoding a 303,476,736 byte divx .avi and from the math it should be around 21 MB I guess. Looks like it's gonna take a while, I'll post the results and any other findings in a comment later on.
- tokachu, on 10/12/2007, -9/+15Read this very carefully:
"With this they can reduce a 23mb 30fps video down to 1519bytes." (Note the "1519 BYTES")
Flag as inaccurate, please. Mathematical errors and typographical mistakes are offenses that cannot be forgiven, and should be considered for the death penalty. - mrASSMAN, on 10/12/2007, -0/+61519 bytes is about 1.48KB (about the size of the tiny avatar to the left of my name.)
- szelij, on 10/12/2007, -1/+7Hmm not enough information in the press release. How fast is the decompression rate? If it can be played on hand-held things i would think it's pretty fast. If so, watch out for an open source version of this technology...
- harrisbradley, on 10/12/2007, -0/+6again, another nominee for best comment of the year
- wscott, on 10/12/2007, -2/+81519 bytes is out of context:
EuclidVision compressed a 23mb 30fps reference video down to 1,519 bytes (15,168:1 compression ratio, 3.56 Kbps bandwidth). MPEG-4 H.264 compressed the same video at similar quality to 8,518 bytes (2,705:1 compression ratio, 19.96 Kbps). Given that MPEG-4 is 50% more efficient than MPEG-2, the 5.6 times improvement over MPEG-4 represents an implied 740% improvement over MPEG-2.
Do the math, the video was only 0.4 seconds long. They also seem to be claiming the high gains are for videos with very little movement. ie.. talking heads - fearlessfrog, on 10/12/2007, -0/+6||||
oo
```
That's me on a video phone using this new compression. It only takes 13 bytes and it even includes my beard!! - ahhell, on 10/12/2007, -0/+5I wonder if this will work with the Phantom? That would be sweet.
- mrASSMAN, on 10/12/2007, -1/+6hmm, i need to spread the news.. maybe i'll submit your comment to digg!
- florin, on 10/12/2007, -0/+5It's definitely either a typo, or a brain fart by the tech writer, or even something worse such as intentional misleading. Hard to tell at this point.
- jeznav, on 10/12/2007, -0/+5(I'm posting my comment here to save you all that reading down below)
Seriously, people who read only the suface of the article shows that they don't know much...
For some people who cannot take drastic statistical numbers will absolutely not believe it until there is proof of this plausible information.
"23mb 30fps video down to 1519bytes".... Yes the article said it in bytes
"Well how in the world can that be? There's no way you can compress 23mb to 1519 bytes"
"Unless its a blank video"
"Ill just flag this article as inaccurate.." etc, etc.
Well, how many compression algorithims do you know? Run Length Encoding, Dictionary based, Huffman? Obviously this video codec is not using those compression techniques but as stated, Discrete Cosine Transformation and Object Based Compression.
For starters, for example MPEG-4 compression based on, keyframe which the codec interpolates(predict) the next frame based on keyframes. Less frames = smaller size , more keyframes better quality. (I'm not going to talk about data rates.)
Ok as for EuclidVision, their method is based on moving objects, so you only need to capture non-moving object once (such as the background). The problem is how many times are you going to capture the moving object? This is using 2d/3d mesh overlay as it analyze moving points.It then uses another overlay mask which the object get affected by the mesh.....Therefore it does not need to record change in pixels but the mesh in which the pixels will get affected.
So it might turn out to be like this:
based on 160x160
Static pixels (non moving objects) with image compression : around 700 bytes
Moving entity objects with image compression : 500 bytes
Mesh points (say aroud 100 points) : 100 bytes
Tweened motion points data (xy vector xy vector final * 100points) with compression : around 3000 bytes
Total size: 4.3 kb
^ This is an example of how it works so don't take it for accuracy.
Basically, the point is it is not impossible for this video codec to compress a 23mb video to 1513 bytes.
And go read this research paper http://www.ece.gatech.edu/research/labs/MCCL/pubs/dwnlds/Yucel_CSVT97.pdf
P.S. Does anyone remember those 64k demos or that 92kb game that we all thought it was impossible? Just let your minds be curious. - geminitojanus, on 10/12/2007, -0/+5"Yes you can compress a stream of video, without audio, that large down to a few scanty kilobytes using temporal-based object based compression - as this is one part of a research project I am doing. There are IEEE articles that are even more recent which provide far better results than that article, which is dated 1997."
Sure you can. And the minute the person sneezes, your compression ratio goes to hell (along with a scene morph, or an object coming in or leaving a frame), and those very demos back in the late 90's exhibited this fatal weakness to the tee.
Hell, MPEG is built around partial temporal object-based compression; all a "B-Frame" really consists of is the difference between the last I frame and the next I frame (though how this difference is generated, be it motion vectors or image data itself is up to the encoder). The more recently invented compressors (as I assume this one and the one you're working on in your research does) can look "outside of the box" in search of objects (e.g. using a wider macro-block for interframe data generation and/or using the more advanced arithmatic coding systems more recently invented, and thusly can more accurately generate P/B-frames without as many temporal glitches as are often seen in darkness and fades in MPEG-2/4 frames) and/or have highly variable size macro-blocks, along with highly advanced motion comp/prediction and way updated framing models to improve compression ratios without killing performance.
We've got people in another department looking at more granular-type object-frame generation (as part of the standard that defines H.264 says that compressors can look around at up to 32-frames instead of the traditional two frames as allowed by MPEG-2/B-Frames; I highly recommend if you want to be on the most innovative front of video compression that you buy a reference copy of the H.264 standard and read it backwards, forwards, on the toilet and in the car. I've not even digested a third of it yet and some of the things H.264 allows in video compression are over my head and I've been in the field for ages).
Sadly, I'm a bit out of touch with the research aspect of it anymore, as my job has me working more on tweaking the software and the actual computers running the encoding/decoding of the data, but I'd love to be back in school working on the new-age compression systems like this one. Sadly though, I doubt if the work that this company claims or the research you're doing now will prove reasonable, or will have complexity problems which will put them outside of the current realm of computing (as it seems all new compression systems do). - joebrodie, on 10/12/2007, -0/+5Here is the math for those not in the know:
H.264 => 8518 bytes * 8 bits/byte = 68144 bits
68144 * 1 Kb/1024 bits = 66.55 Kilobits (Kb)
66.55 Kb/19.96 Kbps = 3.33 sec - dclowd9901, on 10/12/2007, -0/+4What stinks is the vid quality. The picture in the release is for a cell phone. You can compress anything down to kilobytes, as long as you're willing to lose that quality.
I don't care what it says about no loss to vid quality. We're talking about compression here. Something has GOT to be lost in the encoding. - liquidcoooled, on 10/12/2007, -1/+5dan, I think you will find most of those demos DON'T use 3d hardware, I don't recall such things existing back in the Amiga and dos days. Its not that difficult to do 3d matrix operations in a small amount of assembler.
I do have a clue having been a coder for approximately 20 years and have coded numerous specific compression routines so most likely have a clue.
I could create a CBR compression routine or pattern matching routine which CAN produce very very small compressed result files just like these guys, but once you allow generic inputs the results will go completely to pot. I remember hearing about some super zip compressor which promised similar, but theirs also fell apart once they allowed generic content, the only difference here is that they are using spatial elements instead of sequencial characters.
They talk about object based compression routines, which in my eyes means searching for blocks to convert to graphical elements - "ahhh large block of green" -> "rect(l,t,w,h,green), "ahhh a circle of red" -> circle(x,y,r,red)
They also talk about needing to handle smaller compression blocks - their source video is most likely composed of large flat coloured cartoony elements which would end up similar using practically any compression alg.
A movie compressed in this way could then be further compressed using any of the standard text compressors to further reduce the framesize.
Its not rocketscience, and I don't think they are ***** but I just don't think it will work in the real world. - pwncore, on 10/12/2007, -0/+4When do you plan to release the extended edition director's cut superbit criterion collection boxset?
- EdgarVerona, on 10/12/2007, -0/+4I call *****. There is no way that a 23mb video with any realistic qualities inside of it could be described in just over 1500 bytes. No amount of compression coul possibly make the video compress that much and still contain *any* meaningful data.
I mean, I could write a lossy compression algorithm right now that will compress it to 1 byte. That doesn't mean that the data would in any way resemble what it previously had. - danpsmith, on 10/12/2007, -0/+3A two hour movie is still going to occupy a certain amount for audio. If they invented a new audio codec that does the same ***** for audio, I think that would've been a big story when they invented that. So obviously they haven't, and this press release is probably in range of another fractal compressor.
- sundancekid503, on 10/12/2007, -0/+323mb down to 1519 bytes??? I call bullshirt! I think maybe they meant kilobytes?
Either way that would be nuts and I don't believe it. - inactive, on 10/12/2007, -1/+4"...suddenly have turned into an 'expert' in video compression."
Would I have to be an automotive expert if I knew a car company was full of ***** when they claimed they had a car getting 460% better miles per gallon than the most fuel efficient car on the market? No, I'd just have to have a brain and a little bit of common sense. - oepapel, on 10/12/2007, -1/+4Snakeoil. Some one is looking to generate some buzz before they cash out their shares. Nothing to see here, folks. Move along.
-
Show 51 - 100 of 168 discussions
The Digg Toolbar for Firefox lets you Digg, submit content, and keep track of Digg even when you're not on the Digg site. Download the official 
© Digg Inc. 2009
— Content created and posted by Digg users is