Home Made HD Cinema Cameras - Technical Discussion

[Wayne Morellini]

Radek:
Microsoft are planning Xbox branded computers according to reports, I posted some stuff here a while ago. Your right this must be worrying to the manufacturers.

I think Intel's aliance with Apple and Sony's aliance with IBM finally may bring some quality to market place that put lot pressure on M$. I think both Sony and Apple may have plan and may working on it together and will put squeeze on Billy. The sooner the better.

I agree with you. The thing is Intel is, potentially, a dead duck because of no credible answer to Cell, it is like pitting Workstation against Super computer. So what was Apple thinking. So Intel must have something up their sleeve. I know a few projects that can potentially blow CELL out of the water.

Steve:

Thanks for the info. Good to here FFV1 will do at least 3:1. Interesting that none of the commercial ones can cut it. I still wonder, do any of them eliminate sensor noise to increase compression? What is range compression, sounds similar to an idea I was thinking about years ago?

Ronald:
"what..does it mean ??.."

Well there has been talk of applications including word-processor and Internet, and Keyboards, for a while, but I have not heard of Linux (want OS). Will be camera for it.

Sumix:

http://www.sumix.com/optic/index.html
They have many web sites around the world apparently. I think they are in eastern Europe.

I think info@Sumix.com is their front end, but looks like system crash has caused emails to that address to be lost.

Foveon, good for SD like raw images, but cinema version is coming apparently (read other new threads in forum, do search on my posts).

P# links ?? do you mean PS3, I just read them, interesting.

This post from the link, is most interesting:
http://playstation2-linux.com/forum/...p?msg_id=48156

About Sony's/Toshiba/IBM's plans to open up full specs to encourage Open Source development to get the Cell supported and used for new products.

Originally Posted by Ronald Biese

I do guess it comes out first in Japan
do you have friends there to order one, would like to get one.......
..
Other PC Operating Systems can run too, such as Windows and Tiger (Max OS X 10.4), if the publishers want [them] to do so. Maybe a new OS might come out.
sic

They do want to release things worldwide but often manufacturing volume conspires against them. Grey importers are the best bet, but, from memory, Sony is trying to close that down too.

Yes, I was hoping for an emulator to run OSX on Xbox360 or PS3.


Thanks

Wayne.

[Wayne Morellini]

Thanks to Mike from hdforindies for posting this on his site:
http://www.blackmagic-design.com/sup...asp?techID=102

Sony has released a laptop with 1920*1080 capable screen, processor speed not the fastest though.

Wayne.

[Radek Svoboda]

Sony HC1 has uncompressed HD output.

Radek

[Steven Mingam]

Originally Posted by Wayne Morellini

Steve:

Thanks for the info. Good to here FFV1 will do at least 3:1. Interesting that none of the commercial ones can cut it. I still wonder, do any of them eliminate sensor noise to increase compression? What is range compression, sounds similar to an idea I was thinking about years ago?

It does 2.7:1 on a very hard 1080p video sample here, at 4fps (crappy celeron 2ghz). Should be twice slower than HuffYUV. Range coding is a sort of Arithmetic Coding, but twice faster, for a 0.01% loss in compression, more info :
http://www.arturocampos.com/ac_range.html
and http://www.data-compression.info/Algorithms/RC/
It's believed to be patent free, too.

[Wayne Morellini]

Thanks for the compression site link, that should be interesting. I am surprised it is so slow (but for those sorts of compression ratio it should not surprise me). Not feeling well enough for reading at the moment, but from the first paragraph, on range compression, it doesn't seem, related to my scheme. I was thinking where you get a stream, or area and work out it's max and min value, and the max and min value that would give the max compression on the following: A new pixel value (or sub stream or sub area) is expressed as a value from a range of values (offset to start from zero) any exception to this value then goes through an exception handling procedure to give the alternative value or string/area of values, or the sub streams/areas are broken up to contain the value in that range. The whole idea is that if the consecutive value (pixel, stream or area) has a 4 bit change in value, just record 4 bits instead of 8 bits. This can be from an absolute starting value or a changing value dependent on the last value, or predetermined position etc). For a face, or a green leaf, much of the detail (except shard, marks or highlights) then could be encoded in this way producing maybe nearly 2.1 compression before any other compression method is applied. It is only a part solution.

A lot of my stuff depends on integer for speed (except for transformations, irregular area definition, ramping, and the more advanced features that maybe touched upon by the subsequent Intel 3D from 2D photo stuff). I know many of these things might already touched upon in the compression field, but nice to think. Sorry if this explanation is a bit clumsy, off color today.

Have your heard of a compression platform called Adams Platform (probably from 1995) it was written very quickly by a video person in Australia to feed a video wall? It was producing around 1000:1 compression ratio. I never heard about how it worked?

Radek:
Have you found out it is uncompressed from an review article?


Thanks

Wayne.

[Radek Svoboda]

It's right on Sony U.S. site, in camera pages; they even sell fancier cable for that port in accessory pages.

[Wayne Morellini]

I looked before, it mentioned component but not about compression. We wait and see. No manual iris or shutter, bummer! I can get the auto version on $100 toy camera, hardly a feature, pretty cheap of them!

[Steven Mingam]

Originally Posted by Wayne Morellini

Have your heard of a compression platform called Adams Platform (probably from 1995) it was written very quickly by a video person in Australia to feed a video wall? It was producing around 1000:1 compression ratio. I never heard about how it worked?

With i little googling i found this : http://www.theage.com.au/articles/20...&oneclick=true

looks like your Adams Platform was just way to gain $$ during those crazy days... Sound like a french company, I2BP, who claimed HD quality at 2Kb/s ... and didn't even have the quality of mpeg2 when broadcasting on a 10Mbits LAN...
[edit] nice link : http://datacompression.info/IncredibleClaims.shtml [/edit]


Btw, your compression scheme sound like differential coding, and a bit too arithmetic coding, but i'm a little lost in your explaination ;)

And for your comment about speed, don't forget i'm encoding a 1080p @ 25fps video with a first generation celeron at 2ghz, FSB 100Mhz and i'm pretty sure with SD-RAM ! you can't make things worse than that to have the "advantages" of netburst micro-architecture of P4 completly nullified. Celeron should never have see the light, Intel must burn in hell for that.

[Wayne Morellini]

Thanks Steven

Originally Posted by Steven Mingam

looks like your Adams Platform was just way to gain $$ during those crazy days...

The Israeli Intelliegence have, apperently, had 600:1 video imaging in their intelliegence section since the mid 80's. The claim came out within months of the Adams plaform, so I took the possibility a bit more seriously.

"Sound like a french company, I2BP, who claimed HD quality at 2Kb/s ... and didn't even have the quality of mpeg2 when broadcasting on a 10Mbits LAN.."

That's 600 thousand to one, isn't it ;) LIke the one that claimed they could keep compressing the same file repeatedly down to nearly nothing.

"Btw, your compression scheme sound like differential coding, and a bit too arithmetic coding, but i'm a little lost in your explaination ;)"

That OK, my head not that clear today. It is too be expected, that many things I can come up with independently are already in this mature field.

"And for your comment about speed, don't forget i'm encoding a 1080p @ 25fps video with a first generation celeron at 2ghz, FSB 100Mhz and i'm pretty sure with SD-RAM !"

Yes that would slow things quiet a bit. It was more a comment about codec speed though, as quiet often speed critical applications get written in slow C that should have been machine coded.

[Wayne Morellini]

I looked at:
http://www.theage.com.au/articles/20...?oneclick=true

http://www.theage.com.au/articles/20...&oneclick=true

Interesting, not much information on the dis-proof. After thinking about it, I thought that I heard it before 1997, and only the decoder performance was mentioned. At the time I read it the best machines were 386's, I think. I thought it was something like a 320@240 video, as he was comparing it to the VHS images he was replacing. But compression is a lot slower than compression. It is one thing to compress a file of a long video sequence and send it without error correction over a phone line (like he was demonstrating) but compression is much lower to compress live images over the Internet with error correction overhead.

I noted, at the time, that the (newspaper image) seemed to lack detail on the image (I think it might have been a dancer) on a large plain background was plain, which is another way to drive down compression requirements. You can supplement definition by defining the shape (as they do with facial recognition 6000:1+ instead) and transmitting that and filling it in. Even a beach ball on a blue background will increase compression. We can forget the 2KB/s HD codec, I can do it at 30-bit/s, at virtually any resolution, just film a perfect circle and send a draw command to draw it ;)

I think, if you also just restricted the range of values on any details, that maybe transmitted, say to 2-4 bits you also compress further and give a hint at the detail.

So it could be achieved by reducing detail, plane background, outlined shape/edge enhancement/upscaling with no error compression over a phone wire. But on real world images over the Internet, you would fail desperately to get any where near the compression rate. I think that DIVX (I think) had 1000-1500 to one in the early 90's for DVD resolution images over the network, so it is possible, but a quality compromise.

On the Age article, to be fair, you don't need to be a good mathematician to come up with a good new system, it is just that mathematicians have been doing this and can't see how to do it without that sort of complexity. It is credible not to run something on others machines also, too easy to steal, erased disk images can still be read after many overwrites. If it was me, I would probably had provided the money, get them to pick a store out of a phone book by random, drive over there with them and buy a machine, install and demonstrate, and keep the machine.

[Wayne Morellini]

http://news.spong.com/detail/news.asp?prid=8840

Even though the NVIDIA/Sony Hybrid chip might be ahead of schedule, how often have I heard of delays of European release due to complexity of manufacture, and this is the most complex personal computer yet.


Interesting to note ATI is claiming the, approx half the power, Xbox360 is the most advance thing you'll see in the next few years and double what's out there. Are they in self denial, you may ask. Well the interesting thing is that the xbox360 has unified shader structure (where shader and vertex units are combined in one processing model) something I have been waiting for a number of years. If it is what I suspect, it is the long horn combined architecture, that acts a bit more like a general purpose processor. This would be a extremely good thing as far as image processing and compression goes, as these functions could be programmed on graphic chip with minimal processor involvement. This is unlike the present day where the CPU has to keep feeding the shader pipelines to get some benefit out of it, that also may produce some timign problems. ATI claims 100% performance compared to I think it was 50-70% for the old model. This might actually draw real life performance (for us) closer to the Playstation 3. But there is more than enough power many times over as it is, but for programming it might be a bit easier (and closer to future PC/MAC graphic chip models.

http://news.spong.com/detail/news.asp?prid=8829

It is interesting that NVIDIA would not have access to this for the PS3 (but it is a MS technology) so we will wait to see.

[Régine Weinberg]

We will see. Nice thing Sony will ship a disk with a Linux on it. For Linux there is NLE Software, Compositing software, all is there sometime not so blown up as Maya or Shake or FCP but Shrek was made with, Nadagaskar is made with and so on we have to see.

[Wayne Morellini]

..........

[Régine Weinberg]

maybe vented a zillion time
there are CMOS out doing 12.5 fps with a lot af resolution.
There are still cams out using them.
Why not using 2 of such Cmos.
One takes all pair one all impair images.
using35 mm optics.
Fullframe format
or two Foveons
The Steve Nordhauser package is 10K$ and Bayer
and not portable
so crazy but

[Wayne Morellini]

Hello Ronald

Sounds good. I'm more interested in tweaking a 12.5 fps camera to do 24/25 fps.

I've finally got around to starting research on the foveon based Polaroid and it doesn't have a raw VGA video mode, just Mpeg, so I gave up. I was thinking of researching to see if you could program the Sensor to offset the read out of each of the binned/unbinned colour planes from each other, to produce 1/3 rd pixel shift. If it was possible, and I could get a dev kit, I could very cheaply make a 1080p+ RAW camera. 640*480 becomes 1920*1440. Lossless/visually lossless compressibility would then become higher over Bayer. So 3:1 or 6:1 should be possible. The data rate would be much lower than 720p Bayer camera for 1080p. With compression it becomes closer to 5 MB/s. This is the future I think instead of Bayer for single chip. I don't know the exact problems of 1/3 rd pixel shift compared to native 3 chip, but I imagine it is much better than Bayer in many situations. Unfortunately due to all that is happening here I haven't got around to researching it. You can understand why I was so quiet about this.

I am currently thinking about (though haven't got time to do it) about a new lossless compression scheme that would get the pixel shifted 1080p down inside a HDTV channel at a fraction of the processing cost (certainly with cineform you can already get visually lossless just about in a channel, which compared to the quality of the HDTV transmission stuff I already see, would be a "cinematic" improvement).

--------------------------------------------------------------------------
Any Patent and intellectual property rights in this I reserve for myself, so now others can't patent it.