4:4:4 12-bit Uncompressed DVX100

Mark Grgurev · October 2nd, 2004, 10:08 PM

I did a little test in Photoshop for the fun of it and I realized that the XL2 chip can produce a High Definition signal. Pixel shift supposedly increases resolution by 150%. well I took a 853x480(NTSC widescreen) frame grab from the XL2 and increased its size by 150 % and i got a perfect 1280x720!!! With a Pal Xl2, it would be 1536x864, thats only 25% less resolution than 1080p!!!! Of course if you really want to get crazy and put an anamorphic adapter on it for Cinemascope, you get 2016x864. Besides the fact that there is a Pal and NTSC version of the chip, this leads me to believe the XL2 was designed to be HD.

Wayne Morellini · October 3rd, 2004, 06:33 AM

Mark, I don't know what process feature your using to get 150%, but to take advantage of the pixel shift you will need to get the raw frames (seperate with each pixel) for red green and blue and play with those mathematically to see what you can get. If Juan has posted colour seperated frames people you could test it for the 100 first.

I had a thread a while ago about getting HD on a XL2 by pixel shift. The XL2 has a software developers kit. If it allows you to program raw frames out (if it was really good it would) then you could do that.

I think it is all very good idea, as these cameras are going to be cheap come the time the true HD one come doiwn in price, and if you can do it the quality (4:2:0) shouldn't be too far off the HD ones (pluss the firewire hard drive to record it).

Mark Grgurev · October 3rd, 2004, 11:22 AM

I wasn't taking advantage of pixel shift, I just heard that pixel shift boosts resolution by 150%. Since the raw images from the Xl2 will at least be the size of a regular DV image , I just bicubicly uprezzed the footage to see what size NTSC Widescreen footage should be with 150% more resolution.

Thomas Smet · October 3rd, 2004, 07:56 PM

The problem with dv tape I think is that even with the XL2 having chips of that size isn't the rez lowered to 720 x 480 to fit on tape? Then once in the computer the anamorphic 720 frame is pulled back up to 853? Or does the XL2 record a different type of dv frame?

That means even though the XL2 shoots a really nice sharp 16x9 the pixel count is still a 720 x 480 anamorphic frame once captured.

The best way to get HD from the XL2 would be to use an anamorphic adapter at a 1.5 to 1 scale and second to figure out a way to activate the full 720 V pixels of the chips.

This would then give you 1440 x 720 non-square raw pixels. You could then turn those into square pixels to get 1280 x 720.

The interesting thing about this is that even though the chips are only 960 pixels wide and you have to blow up to get 1280 the new SONY HDV camera does the same thing. The SONY HDV camera uses chips that are 960 x 1080. That means the chips are blown up by 2X to get 1920 pixels. Well they are first blown up to 1440 but that is anamorphic. The SONY HDV starts with high H pixel count but blown up V pixels. The XL2 could be made to do basically the same thing but only to get 1280 x 720.

Of course the odds of getting a 1.5 anamorphic lens and figuring out if those "non-fired" pixels can be used are very slim so I wouldn't hold my breath for it to happen.

Juan P. Pertierra · October 3rd, 2004, 08:23 PM

To Brett:

My device does not have an internal hard disk, it's an interface to a PC or a external hard disk, so you couldn't interface it to DV like you say. However, the device does have user-set options which allow the recording of a stream which is approximately the same size as a DV stream...it is NOT the DV stream created by the camera though.

Juan

Mark Grgurev · October 3rd, 2004, 09:27 PM

To Thomas:

If your method did work than the pixel shift with give you a full 1920x1080. The pal version would get you 2048x1152. It wouldn't be as good as the F950 but I won't complain.

Juan P. Pertierra · October 3rd, 2004, 09:31 PM

Remember that I have actually posted a 3 layer photoshop file in which the green layer is pixel shifted. You can try your methods with that file and see if it really works.

http://expert.cc.purdue.edu/~pertierr/shift_work.psd

Chris Rubin · October 3rd, 2004, 10:25 PM

Good to hear from you, Juan. I already thought that the Men in Black from Canon or Sony had kidnapped you.

Thomas Smet · October 5th, 2004, 10:28 PM

I thought this might be helpful to us for deciding which codec to use for our raw video.

This guy compared a lot of the top video codecs including 4:1:1, 4:2:2, and 4:4:4.

He compared and tested both the Bitjazz Sheervideo codec and the microcosm codec. While Microcosm gives an insane amount of lossless compression it is slow. Sheervideo doesn't save as much space but is faster than realtime so may even work to encode directly from camera.

This site is also a good place to check out other types of codecs and see how bad DV really is compared to other formats.

The only bad thing is that only Mac codecs are tested and none for the PC yet.

http://www.onerivermedia.com/codecs/index.htm

Brad Abrahams · October 6th, 2004, 06:06 AM

Sheervideo works fine for the PC, I am a beta tester. Microcosm also works on the PC as well

Unfortunately, Sheervideo and Microcosm are not free, so I do not think this option is for everybody.

Juan, it may be a good idea to contact the Cineform guys (www.cineform.com) who make Aspect and Connect HD. They've developed an extremely efficient and fast codec that encodes 10-bit avi files with HD resolution. It encodes and playback faster than realtime without any fancy hardware. You may be able to work out some sort of bundle deal, including the codec with the hardware modification.

Thomas Smet · October 6th, 2004, 10:47 AM

This would be great if we plan on getting HD out of our dvx100 but what about SD?

Even though Sheervideo and Microcosm have a cost I did not think it was to high for people who actually want this type of footage. The only real purpose to this adapter is for high end film or visual effects work. Clearly it wouldn't be used to shoot cousin Billy's Bar Mitzvah.

At the same time people do not have to use these codecs. They could use Tiffs or any other image format they want to for free. Buying one of these codecs is to help deal with the footage better and have better storage. They are just to help and are an option.

People could always use other codecs as well such as animation with slight image loss but that is the tradeoff.

October 6th, 2004, 01:59 PM

i think the whole point of juan's mod is to give us the highest quality output possible from the camera -- which means RAW frames. compression to a codec would just make juan's job more complicated, not mention more expensive.

once we get the raw frames on our hard drives, what codec we use to work with the footage is our problem, and we have scores of options to choose from based on our workstations and workflows. and in terms of user design, juan's mod being only raw output is actually the best approach, because if he were to settle on a particular codec, then i'm sure he'd be alienating some users who prefer a different codec. besides, there is just no getting around the fact that, currently, any good non-lossy codec requires either lots of computation power and/or significant render time.

seriously, we couldn't ask for anything better than a camera that outputs raw pristine frames while SIMULTAMEOUSLY outputting to tape a standardized, low bandwidth format (dv25) for hassle-free offline editing, all in one relatively small package with just one simple firewire tether to a hard drive. there are plenty of other options out there if you want a low or moderately compressed format (dv50, dv100, sonyhd, etc.) that your computer can handle in realtime, and the rental costs for those packages aren't that bad anymore.

Wayne Morellini · October 7th, 2004, 12:05 AM

Over at the digital cinema camera threads (see list at start of technical thread) there has been people eager about doing FPGA based compression solutions, in times past (there was even a short lived FPGA thread). I believe some people, are working on them or do intend too. One company has offered to incorporate such a FPGA design in their cameras for us, if we design it. So as there is some mutal overlap here (single chip bayer is only our present option), you might get some mutal help by combining efforts with them. The Russian company Elphel (??, see home made camera thread) has supplied ccompression FPGA to sourceforge for their own network cameras too (but I think it was a high loss codec, though I don't know exactly what their future plans are). So such a thing could be mounted to a PCI card.

Apart from that there is ways to program onboard DSP's and AGP GPU circuits (only 266MB/s back channel to main/CPU memory. This hasn't been researched yet, but maybe a cheaper way.

We have been looking at low end CPU for aquistion and the data rates for aquistion of bayer 720p and DVX100 4:4:4 is similiar, except for the extra debayering step in the preview. So you should be able to get a good idea of just how low in CPU power you can go from these projects (programming is not yet finished so cpu power requirments will still be dropping). But we hope to eventually have 12cm aquare nano-itxs PC's aquiring the image through PCI capture cards, firewire or Gigabit ethernet.

Cineform is probably the best (costs a nice penny too). There is (among others) Avid's DNxHD codec that goes upto 220Mb/s, but it is limited to 10 bit. Anyway such datarates are very good for video work and some film work (we go upto raw over at the Cinenema threads for highest cinema quality, and certain editing/special effects processing) but on many productions that won't be needed.

Note on interfaces, if you go the route of an external SDI capture board (sorry I haven't read all the thread yet to know your plans) then I hear Heroinewarrior has free Linux 12bit Capture and NLE app, Cinelerra, for SDI input.

Thanks

Wayne.

Jason Rodriguez · October 8th, 2004, 05:57 AM

Quote:

Juan, it may be a good idea to contact the Cineform guys (www.cineform.com) who make Aspect and Connect HD. They've developed an extremely efficient and fast codec that encodes 10-bit avi files with HD resolution.

That sounds like a great idea.

Now I think David Newman is a great guy and very helpful, and his codec is very nice, but why do we have to alienate all the Mac users on this forum (and possible Linux users too)? Especially when I'm sure there's a lot of us who use FCP for editing, especially with the DVX100. I've talked to David many times about Mac support for Cineform, and it's not coming for quite some time due to dependancies for certain libraries, etc. on Windows and it's programming environment.

So again, I have to agree with Jaan, keep it RAW!

Mark Grgurev · October 11th, 2004, 11:09 AM

u done with the website yet, juan?

October 2nd, 2004, 10:08 PM	#1171
Mark Grgurev Regular Crew Join Date: Dec 2003 Location: Fairview,nj Posts: 137	I did a little test in Photoshop for the fun of it and I realized that the XL2 chip can produce a High Definition signal. Pixel shift supposedly increases resolution by 150%. well I took a 853x480(NTSC widescreen) frame grab from the XL2 and increased its size by 150 % and i got a perfect 1280x720!!! With a Pal Xl2, it would be 1536x864, thats only 25% less resolution than 1080p!!!! Of course if you really want to get crazy and put an anamorphic adapter on it for Cinemascope, you get 2016x864. Besides the fact that there is a Pal and NTSC version of the chip, this leads me to believe the XL2 was designed to be HD.

October 3rd, 2004, 06:33 AM	#1172
Wayne Morellini Inner Circle Join Date: May 2003 Location: Australia Posts: 2,762	Mark, I don't know what process feature your using to get 150%, but to take advantage of the pixel shift you will need to get the raw frames (seperate with each pixel) for red green and blue and play with those mathematically to see what you can get. If Juan has posted colour seperated frames people you could test it for the 100 first. I had a thread a while ago about getting HD on a XL2 by pixel shift. The XL2 has a software developers kit. If it allows you to program raw frames out (if it was really good it would) then you could do that. I think it is all very good idea, as these cameras are going to be cheap come the time the true HD one come doiwn in price, and if you can do it the quality (4:2:0) shouldn't be too far off the HD ones (pluss the firewire hard drive to record it).

October 3rd, 2004, 11:22 AM	#1173
Mark Grgurev Regular Crew Join Date: Dec 2003 Location: Fairview,nj Posts: 137	I wasn't taking advantage of pixel shift, I just heard that pixel shift boosts resolution by 150%. Since the raw images from the Xl2 will at least be the size of a regular DV image , I just bicubicly uprezzed the footage to see what size NTSC Widescreen footage should be with 150% more resolution.

October 3rd, 2004, 07:56 PM	#1174
Thomas Smet Trustee Join Date: Mar 2004 Location: Milwaukee, WI Posts: 1,719	The problem with dv tape I think is that even with the XL2 having chips of that size isn't the rez lowered to 720 x 480 to fit on tape? Then once in the computer the anamorphic 720 frame is pulled back up to 853? Or does the XL2 record a different type of dv frame? That means even though the XL2 shoots a really nice sharp 16x9 the pixel count is still a 720 x 480 anamorphic frame once captured. The best way to get HD from the XL2 would be to use an anamorphic adapter at a 1.5 to 1 scale and second to figure out a way to activate the full 720 V pixels of the chips. This would then give you 1440 x 720 non-square raw pixels. You could then turn those into square pixels to get 1280 x 720. The interesting thing about this is that even though the chips are only 960 pixels wide and you have to blow up to get 1280 the new SONY HDV camera does the same thing. The SONY HDV camera uses chips that are 960 x 1080. That means the chips are blown up by 2X to get 1920 pixels. Well they are first blown up to 1440 but that is anamorphic. The SONY HDV starts with high H pixel count but blown up V pixels. The XL2 could be made to do basically the same thing but only to get 1280 x 720. Of course the odds of getting a 1.5 anamorphic lens and figuring out if those "non-fired" pixels can be used are very slim so I wouldn't hold my breath for it to happen.

October 3rd, 2004, 08:23 PM	#1175
Juan P. Pertierra Major Player Join Date: Jul 2003 Posts: 479	To Brett: My device does not have an internal hard disk, it's an interface to a PC or a external hard disk, so you couldn't interface it to DV like you say. However, the device does have user-set options which allow the recording of a stream which is approximately the same size as a DV stream...it is NOT the DV stream created by the camera though. Juan

October 3rd, 2004, 09:27 PM	#1176
Mark Grgurev Regular Crew Join Date: Dec 2003 Location: Fairview,nj Posts: 137	To Thomas: If your method did work than the pixel shift with give you a full 1920x1080. The pal version would get you 2048x1152. It wouldn't be as good as the F950 but I won't complain.

October 3rd, 2004, 09:31 PM	#1177
Juan P. Pertierra Major Player Join Date: Jul 2003 Posts: 479	Remember that I have actually posted a 3 layer photoshop file in which the green layer is pixel shifted. You can try your methods with that file and see if it really works. http://expert.cc.purdue.edu/~pertierr/shift_work.psd

October 3rd, 2004, 10:25 PM	#1178
Chris Rubin Regular Crew Join Date: Jul 2004 Posts: 38	Good to hear from you, Juan. I already thought that the Men in Black from Canon or Sony had kidnapped you.

October 5th, 2004, 10:28 PM	#1179
Thomas Smet Trustee Join Date: Mar 2004 Location: Milwaukee, WI Posts: 1,719	I thought this might be helpful to us for deciding which codec to use for our raw video. This guy compared a lot of the top video codecs including 4:1:1, 4:2:2, and 4:4:4. He compared and tested both the Bitjazz Sheervideo codec and the microcosm codec. While Microcosm gives an insane amount of lossless compression it is slow. Sheervideo doesn't save as much space but is faster than realtime so may even work to encode directly from camera. This site is also a good place to check out other types of codecs and see how bad DV really is compared to other formats. The only bad thing is that only Mac codecs are tested and none for the PC yet. http://www.onerivermedia.com/codecs/index.htm

October 6th, 2004, 06:06 AM	#1180
Brad Abrahams Regular Crew Join Date: Mar 2004 Posts: 65	Sheervideo works fine for the PC, I am a beta tester. Microcosm also works on the PC as well Unfortunately, Sheervideo and Microcosm are not free, so I do not think this option is for everybody. Juan, it may be a good idea to contact the Cineform guys (www.cineform.com) who make Aspect and Connect HD. They've developed an extremely efficient and fast codec that encodes 10-bit avi files with HD resolution. It encodes and playback faster than realtime without any fancy hardware. You may be able to work out some sort of bundle deal, including the codec with the hardware modification.

October 6th, 2004, 10:47 AM	#1181
Thomas Smet Trustee Join Date: Mar 2004 Location: Milwaukee, WI Posts: 1,719	This would be great if we plan on getting HD out of our dvx100 but what about SD? Even though Sheervideo and Microcosm have a cost I did not think it was to high for people who actually want this type of footage. The only real purpose to this adapter is for high end film or visual effects work. Clearly it wouldn't be used to shoot cousin Billy's Bar Mitzvah. At the same time people do not have to use these codecs. They could use Tiffs or any other image format they want to for free. Buying one of these codecs is to help deal with the footage better and have better storage. They are just to help and are an option. People could always use other codecs as well such as animation with slight image loss but that is the tradeoff.

October 6th, 2004, 01:59 PM	#1182
Guest Posts: n/a	i think the whole point of juan's mod is to give us the highest quality output possible from the camera -- which means RAW frames. compression to a codec would just make juan's job more complicated, not mention more expensive. once we get the raw frames on our hard drives, what codec we use to work with the footage is our problem, and we have scores of options to choose from based on our workstations and workflows. and in terms of user design, juan's mod being only raw output is actually the best approach, because if he were to settle on a particular codec, then i'm sure he'd be alienating some users who prefer a different codec. besides, there is just no getting around the fact that, currently, any good non-lossy codec requires either lots of computation power and/or significant render time. seriously, we couldn't ask for anything better than a camera that outputs raw pristine frames while SIMULTAMEOUSLY outputting to tape a standardized, low bandwidth format (dv25) for hassle-free offline editing, all in one relatively small package with just one simple firewire tether to a hard drive. there are plenty of other options out there if you want a low or moderately compressed format (dv50, dv100, sonyhd, etc.) that your computer can handle in realtime, and the rental costs for those packages aren't that bad anymore.

October 7th, 2004, 12:05 AM	#1183
Wayne Morellini Inner Circle Join Date: May 2003 Location: Australia Posts: 2,762	Over at the digital cinema camera threads (see list at start of technical thread) there has been people eager about doing FPGA based compression solutions, in times past (there was even a short lived FPGA thread). I believe some people, are working on them or do intend too. One company has offered to incorporate such a FPGA design in their cameras for us, if we design it. So as there is some mutal overlap here (single chip bayer is only our present option), you might get some mutal help by combining efforts with them. The Russian company Elphel (??, see home made camera thread) has supplied ccompression FPGA to sourceforge for their own network cameras too (but I think it was a high loss codec, though I don't know exactly what their future plans are). So such a thing could be mounted to a PCI card. Apart from that there is ways to program onboard DSP's and AGP GPU circuits (only 266MB/s back channel to main/CPU memory. This hasn't been researched yet, but maybe a cheaper way. We have been looking at low end CPU for aquistion and the data rates for aquistion of bayer 720p and DVX100 4:4:4 is similiar, except for the extra debayering step in the preview. So you should be able to get a good idea of just how low in CPU power you can go from these projects (programming is not yet finished so cpu power requirments will still be dropping). But we hope to eventually have 12cm aquare nano-itxs PC's aquiring the image through PCI capture cards, firewire or Gigabit ethernet. Cineform is probably the best (costs a nice penny too). There is (among others) Avid's DNxHD codec that goes upto 220Mb/s, but it is limited to 10 bit. Anyway such datarates are very good for video work and some film work (we go upto raw over at the Cinenema threads for highest cinema quality, and certain editing/special effects processing) but on many productions that won't be needed. Note on interfaces, if you go the route of an external SDI capture board (sorry I haven't read all the thread yet to know your plans) then I hear Heroinewarrior has free Linux 12bit Capture and NLE app, Cinelerra, for SDI input. Thanks Wayne.

October 11th, 2004, 11:09 AM	#1185
Mark Grgurev Regular Crew Join Date: Dec 2003 Location: Fairview,nj Posts: 137	u done with the website yet, juan?