Dan Hudgins
Well-known member
Thanks,
I got frustrated trying to get anything to not have block artifacts on Vimeo, and even if I do to some extent, the H.264 is a pale shadow of what the True RAW cameras can capture. I wish people could come over and look at the various cameras, BMDCC (tm), Aaton (tm), KineRAW-S35 (tm), Acam dII (tm), Kinor-2K (tm), and others like T3i and RED ONE, on my 2K 2048x1535 CRT grading monitor so they could see better how they compare under conditions that let you see what is in the raw data. You can't tell anything from a JPG still, I have done tests even with quality set to 100% JPG show losses on computer generated images, those are there is camera images as well but maybe less obvious because of the noise dithering, noiseless images are a kind of paradox, you need noiseless images for interframe compression formats like H.264 that only have a full refresh rate of like 8fps (force keyframe etc.) so if you show grain it starts and stops moving and looks gastly not like 24fps grain in a film projection.
Cineform (tm) blurs things a bit because its wavelet compression so that makes the single pixel digital grain look a bit film-like because film grains are not within a single pixel at 2K resolution, more like spread over four to nine pixels for color negative high speed film stocks. At least Cineform (tm) only compresses within each frame rather than in time, so the grain moves at 24fps like film projection would.
With DNG you get crisp sharp grain 1:1 for each pixel, the only blur is from the OLPF filter which is needed to avoid chroma moire. You cannot get the color out of a Bayer filter if light only falls on a single pixel, so you NEED to blur the light over four pixel groups at least in order to get the color information to have accurate hue and saturation. The luma information can be sharpned to counter the blur from the OLPF, but the chroma should not be sharpned as that just revereses the needed job of the OLPF in extracting the color information.
As far as I can tell the Kinor-2K (tm) does not have an OLPF in the frames I got, and shows very strong color artifacts.
Acam dII (tm) seemed to think that they did not need one because the CCD has a larger fill factor, I'm not sure if they are using one now or not, I suggested to them that it was not a choice.
BlackMagic (tm) seems to have a under blur on their OLPF, but processed in my de-Bayer program I was able to filter some of the chroma moire out because of the over-sample when reducing to HD size.
Likewise in the Aaton (tm) the larger oversample helps when going down to HD size, and it looks quite good at that reduction and lower ISO (CCD sensor), but I don't think anyone in the indy film DCC camera buying range will be able to afford one to own?
KineRAW-S35 (tm) does not have much oversample at 2048x1080 for HD use, or 1:1 pixel for DCP use, but it has a custom made OLPF filter that helps control the moire and aliasing well, although at 1:1 pixel if the sharpen is set to high you can counter the OLPF like you could in any other camera, but you have a good adjustment range on the luma sharpen because the OLPF they are using was matched to the larger effective pixel size in the sensor.
SI-2K (tm) I think lets you shoot with or without the OLPF, but that is not a good idea if you have lenses that can resolve well, they have the issue of a 2/3" sensor so good lenses for that format are not as simple an issue as for the KineRAW-S35 (tm) where you can use many SLR and DSLR lenses as well as most 35mm movie camera lenses, I have mounted Mitchell NC, Eyemo, 2709, and OCT-19, as well as M42 so far and they all seem to work well. It was a supprise that a Russian 35mm OCT-19 lens seems to be a bit sharper than a Cook Speed Panchro Series II 32mm even taking into account that the details are a bit smaller at the 32mm focal length. For real theratical shooting, the KineRAW-S35 (tm) has these advantages:
1) The sensor size matched 35mm projection apature about so you shoot what you would see in a movie theatre with the same lens.
2) You can use real 35mm movie lenses with manual iris and follow focus gears.
3) The sensor has analog gain override that gives wide control over the head-room stops as well as the ISO, from about 80 to 160000. Although the higher speeds are to grany for filmmaking, speeds up to about 28000 seem to have been usable with NR applied in post, and speeds of 1280 to 5120 are quite usable, if I shot T500 film pushed two stops it might look worse at 2000 than the KineRAW-S35 (tm) looks at 2560, perhaps, it depends on if the film scans are de-grained as part of the "dust busting" applied to film scans. In both cases film gets de-grained and digital gets NR reduced, so to be fair you need to take into account that both film and digital images are processed before you see them on the Blu-ray (tm).
4) The BOTH recording mode of DNG and Cineform (tm) at the same time lets you compare the RGB result (DPX or TIF) from both recordings, the Cineform (tm) are log encoded and the DNG are sensor linear so have no lost tones of histogram gaps. The MOV Cineform (tm) play back imedately and can be edited without transcoding when the Cineform (tm) codec is loaded. If you want different results from what the Cineform (tm) codec can produce you can then de-Bayer the exact same frames from the DNG in any program you want to use that supports DNG, and are not limited to the Adobe (tm) SDK's or the Cineform (tm) codec's look.
5) The sub-PL mount on the KineRAW-S35 (tm) alows you to swap lens mounts without the need to re-adjust the Flange to Focal Distance, useful if you have movie and SLR lenses already and want to change lens mounts in a fast and simple way while shooting. So far I have not noticed any dust problems when changing the lens mounts, even outdoors and shooting at f/11 or maybe smaller.
6) The camera has built in waveform monitor on the viewfinder screen for quick and accurate exposure if you know what its there for. It can monitor both sensor RAW as well as monitor LUT adjusted.
7) The camera has monitoring LUT for, KineCOLOR (tm) that gives full range output for maximum quality with a bit softer results, KineLOG (tm) to match 35mm film scans using standard Cineon (tm) specs for the placement of midtone and other key tonal values, those are ISO adjusted so no grading is required in film log workflows if the camera exposure was set right on the waveform and histogram, and Kine709 (tm) that outputs broadcast contrast and saturation 3D-LUT that give ready to use footage for transmission without grading if the exposure was set right when filmming. There are also overrides in the 3D-LUT output from the monitoring for your worflow that can convert KineCOLOR (tm) to softclip Cineon (tm) film scan log results eliminating the need to softclip the KineLOG (tm) results in your workflow, and you can get Rec.709 output with mid tone adjusted for video standard (a little darker) or film standard to lighten skin tones of people with darker complections for Broadcast use without grading. At the same time you can de-select the 3D-LUT to disable the monitoring matching in your workflow and make any corrections you like from the non-ISO adjusted Cineform (tm) (that is where some of the faults in the samples you may have seen can happen, as the Cineform (tm) is an absolute recording NOT ISO adjusted, to get ISO adjusted results the 3D-LUT saved in each shot folder at the time of shooting needs to be loaded and active in the Cineform (tm) codec to convert the absolute signal levels to ISO adjusted signal levels, then grading can be done from the ISO adjusted signal levels as it would be from a film scan of film shot at its rated ISO.) The DNG have no curve they are 100% sensor linear data, but there is a second 3D-LUT saved that can be used to mate the DNG data to the other 3D-LUT that matches the monitoring LUT used at the time of shooting, again to get color matching the person making the graded footage needs to load the 3D-LUT generated at the time of shooting or the conversion will be ad-hock and not ISO adjusted by calibration to the waveform and histogram. You have the ability to do as you like, or you can use calibrated results, but you need to understand which is which in planning the workflow, hoperully people that purchase the camera will have someone explain that in a clear way so they understand the reasoning behind how the camer works. As well as understanding the differences between setting the wavefor to sensor signal levels (RAW) or to monitoring signal lavels which can be several stops different (Post). Again its better to have a camera that lets you view BOTH RAW and POST waveforms, so you can judge the highlight expaoure at higher ISO when you cannot see them on the small LCD monitors, but as the camerman you need to KNOW why those features were built into the design to take advantage of them so you can quickly set the exposure and gain to get optimized image recordings.
8) In the KineRAW-S35 (tm) the color balance of light type, Tungsten, Florescent,, LED, skylight, sunset, etc, are done in a special way that other camera makers may not use, and this special way of adjusting the color balance keeps the signal strong on all three color channels making the 12bit data 12bits no matter what the light or color balance is, that can be different from Digital Cinema Cameras that always record at native sensor balance and so you can need to clip the top bits off to get color balance, such as at very low K values only having maybe 9 bits in the blue, with KineRAW-S35 (tm) you get full 12bits ADC output all the time when you set the K and light type to the light you are filming under. Likewise, you can set the K and light type to values other than what is on the set to get warm or cool effects in the camera and reduce the need to grade later and add histogram gaps to your images, and even if you adjust the color balance a bit in post, the changes can be less extream so you end up with smoother image tones and fewer histogram gaps in some cases because the KineRAW-S35 (tm) signal is stronger to start with on all three color channels. The camer can also record at native sensor balance, equal minimum gain, but there is no advantage to doing that as digital gain needed for color balance, like in other cameras, will cause gaps in the tonal range, the KineRAW-S35 (tm) can avoid digital gain in post and give smoother tones for its given ADC bits when used as setup by the supplied light type and K balance tables that the firmware can load (you can add additional settings that will show up in the button menus if you want as well for custom color balance of the sensor (not all settings are valid, but you can see that on the monitor before shooting)).
A digital cinema camera, is not ment to be "fool proof" in that move cameras have full MANUAL control for a reason, there may be more "fools" in the world than people that are skilled cinematographers, but skilled cinematographers even ones that do not have big budgets should have acess to a cinematic tool that gives them 100% control over exposure and sensor gain and records 100% of the sensor data so they can take best advantage of what they are recording without losses, losses can be decided on later and reversed if you record everything the sensor can put out.
The more people that have access to a manually controlled camera in a price range that is affordable, the more younder filmmakers there will be that will have the skills of a cinematographer into the future, they may make some mistakes, but so did people shooting film, and learning from mistakes gives you the ability to control the light to bring the vision to the screen. Replacing everything with CGI and running the camera on automatic under bad lighting do not make better movies just cheeper ones.
==
I working on a Linux Ubuntu port of my de-Bayer program, that can maybe reduce the cost of the render farm for final grading to 48bpp DPX frames. That issue is more important for 4K uncompressed DI than 2K because you have 4K the data and time involved and so would need 4x the number of render computers, using the free OS Ubuntu on junk computers can reduce the cost of setting up a render farm without loss of image quality.
Hopefully someday DCP will be uncompressed as well so people who sit in the front row can see what the sensor captured as well...
I got frustrated trying to get anything to not have block artifacts on Vimeo, and even if I do to some extent, the H.264 is a pale shadow of what the True RAW cameras can capture. I wish people could come over and look at the various cameras, BMDCC (tm), Aaton (tm), KineRAW-S35 (tm), Acam dII (tm), Kinor-2K (tm), and others like T3i and RED ONE, on my 2K 2048x1535 CRT grading monitor so they could see better how they compare under conditions that let you see what is in the raw data. You can't tell anything from a JPG still, I have done tests even with quality set to 100% JPG show losses on computer generated images, those are there is camera images as well but maybe less obvious because of the noise dithering, noiseless images are a kind of paradox, you need noiseless images for interframe compression formats like H.264 that only have a full refresh rate of like 8fps (force keyframe etc.) so if you show grain it starts and stops moving and looks gastly not like 24fps grain in a film projection.
Cineform (tm) blurs things a bit because its wavelet compression so that makes the single pixel digital grain look a bit film-like because film grains are not within a single pixel at 2K resolution, more like spread over four to nine pixels for color negative high speed film stocks. At least Cineform (tm) only compresses within each frame rather than in time, so the grain moves at 24fps like film projection would.
With DNG you get crisp sharp grain 1:1 for each pixel, the only blur is from the OLPF filter which is needed to avoid chroma moire. You cannot get the color out of a Bayer filter if light only falls on a single pixel, so you NEED to blur the light over four pixel groups at least in order to get the color information to have accurate hue and saturation. The luma information can be sharpned to counter the blur from the OLPF, but the chroma should not be sharpned as that just revereses the needed job of the OLPF in extracting the color information.
As far as I can tell the Kinor-2K (tm) does not have an OLPF in the frames I got, and shows very strong color artifacts.
Acam dII (tm) seemed to think that they did not need one because the CCD has a larger fill factor, I'm not sure if they are using one now or not, I suggested to them that it was not a choice.
BlackMagic (tm) seems to have a under blur on their OLPF, but processed in my de-Bayer program I was able to filter some of the chroma moire out because of the over-sample when reducing to HD size.
Likewise in the Aaton (tm) the larger oversample helps when going down to HD size, and it looks quite good at that reduction and lower ISO (CCD sensor), but I don't think anyone in the indy film DCC camera buying range will be able to afford one to own?
KineRAW-S35 (tm) does not have much oversample at 2048x1080 for HD use, or 1:1 pixel for DCP use, but it has a custom made OLPF filter that helps control the moire and aliasing well, although at 1:1 pixel if the sharpen is set to high you can counter the OLPF like you could in any other camera, but you have a good adjustment range on the luma sharpen because the OLPF they are using was matched to the larger effective pixel size in the sensor.
SI-2K (tm) I think lets you shoot with or without the OLPF, but that is not a good idea if you have lenses that can resolve well, they have the issue of a 2/3" sensor so good lenses for that format are not as simple an issue as for the KineRAW-S35 (tm) where you can use many SLR and DSLR lenses as well as most 35mm movie camera lenses, I have mounted Mitchell NC, Eyemo, 2709, and OCT-19, as well as M42 so far and they all seem to work well. It was a supprise that a Russian 35mm OCT-19 lens seems to be a bit sharper than a Cook Speed Panchro Series II 32mm even taking into account that the details are a bit smaller at the 32mm focal length. For real theratical shooting, the KineRAW-S35 (tm) has these advantages:
1) The sensor size matched 35mm projection apature about so you shoot what you would see in a movie theatre with the same lens.
2) You can use real 35mm movie lenses with manual iris and follow focus gears.
3) The sensor has analog gain override that gives wide control over the head-room stops as well as the ISO, from about 80 to 160000. Although the higher speeds are to grany for filmmaking, speeds up to about 28000 seem to have been usable with NR applied in post, and speeds of 1280 to 5120 are quite usable, if I shot T500 film pushed two stops it might look worse at 2000 than the KineRAW-S35 (tm) looks at 2560, perhaps, it depends on if the film scans are de-grained as part of the "dust busting" applied to film scans. In both cases film gets de-grained and digital gets NR reduced, so to be fair you need to take into account that both film and digital images are processed before you see them on the Blu-ray (tm).
4) The BOTH recording mode of DNG and Cineform (tm) at the same time lets you compare the RGB result (DPX or TIF) from both recordings, the Cineform (tm) are log encoded and the DNG are sensor linear so have no lost tones of histogram gaps. The MOV Cineform (tm) play back imedately and can be edited without transcoding when the Cineform (tm) codec is loaded. If you want different results from what the Cineform (tm) codec can produce you can then de-Bayer the exact same frames from the DNG in any program you want to use that supports DNG, and are not limited to the Adobe (tm) SDK's or the Cineform (tm) codec's look.
5) The sub-PL mount on the KineRAW-S35 (tm) alows you to swap lens mounts without the need to re-adjust the Flange to Focal Distance, useful if you have movie and SLR lenses already and want to change lens mounts in a fast and simple way while shooting. So far I have not noticed any dust problems when changing the lens mounts, even outdoors and shooting at f/11 or maybe smaller.
6) The camera has built in waveform monitor on the viewfinder screen for quick and accurate exposure if you know what its there for. It can monitor both sensor RAW as well as monitor LUT adjusted.
7) The camera has monitoring LUT for, KineCOLOR (tm) that gives full range output for maximum quality with a bit softer results, KineLOG (tm) to match 35mm film scans using standard Cineon (tm) specs for the placement of midtone and other key tonal values, those are ISO adjusted so no grading is required in film log workflows if the camera exposure was set right on the waveform and histogram, and Kine709 (tm) that outputs broadcast contrast and saturation 3D-LUT that give ready to use footage for transmission without grading if the exposure was set right when filmming. There are also overrides in the 3D-LUT output from the monitoring for your worflow that can convert KineCOLOR (tm) to softclip Cineon (tm) film scan log results eliminating the need to softclip the KineLOG (tm) results in your workflow, and you can get Rec.709 output with mid tone adjusted for video standard (a little darker) or film standard to lighten skin tones of people with darker complections for Broadcast use without grading. At the same time you can de-select the 3D-LUT to disable the monitoring matching in your workflow and make any corrections you like from the non-ISO adjusted Cineform (tm) (that is where some of the faults in the samples you may have seen can happen, as the Cineform (tm) is an absolute recording NOT ISO adjusted, to get ISO adjusted results the 3D-LUT saved in each shot folder at the time of shooting needs to be loaded and active in the Cineform (tm) codec to convert the absolute signal levels to ISO adjusted signal levels, then grading can be done from the ISO adjusted signal levels as it would be from a film scan of film shot at its rated ISO.) The DNG have no curve they are 100% sensor linear data, but there is a second 3D-LUT saved that can be used to mate the DNG data to the other 3D-LUT that matches the monitoring LUT used at the time of shooting, again to get color matching the person making the graded footage needs to load the 3D-LUT generated at the time of shooting or the conversion will be ad-hock and not ISO adjusted by calibration to the waveform and histogram. You have the ability to do as you like, or you can use calibrated results, but you need to understand which is which in planning the workflow, hoperully people that purchase the camera will have someone explain that in a clear way so they understand the reasoning behind how the camer works. As well as understanding the differences between setting the wavefor to sensor signal levels (RAW) or to monitoring signal lavels which can be several stops different (Post). Again its better to have a camera that lets you view BOTH RAW and POST waveforms, so you can judge the highlight expaoure at higher ISO when you cannot see them on the small LCD monitors, but as the camerman you need to KNOW why those features were built into the design to take advantage of them so you can quickly set the exposure and gain to get optimized image recordings.
8) In the KineRAW-S35 (tm) the color balance of light type, Tungsten, Florescent,, LED, skylight, sunset, etc, are done in a special way that other camera makers may not use, and this special way of adjusting the color balance keeps the signal strong on all three color channels making the 12bit data 12bits no matter what the light or color balance is, that can be different from Digital Cinema Cameras that always record at native sensor balance and so you can need to clip the top bits off to get color balance, such as at very low K values only having maybe 9 bits in the blue, with KineRAW-S35 (tm) you get full 12bits ADC output all the time when you set the K and light type to the light you are filming under. Likewise, you can set the K and light type to values other than what is on the set to get warm or cool effects in the camera and reduce the need to grade later and add histogram gaps to your images, and even if you adjust the color balance a bit in post, the changes can be less extream so you end up with smoother image tones and fewer histogram gaps in some cases because the KineRAW-S35 (tm) signal is stronger to start with on all three color channels. The camer can also record at native sensor balance, equal minimum gain, but there is no advantage to doing that as digital gain needed for color balance, like in other cameras, will cause gaps in the tonal range, the KineRAW-S35 (tm) can avoid digital gain in post and give smoother tones for its given ADC bits when used as setup by the supplied light type and K balance tables that the firmware can load (you can add additional settings that will show up in the button menus if you want as well for custom color balance of the sensor (not all settings are valid, but you can see that on the monitor before shooting)).
A digital cinema camera, is not ment to be "fool proof" in that move cameras have full MANUAL control for a reason, there may be more "fools" in the world than people that are skilled cinematographers, but skilled cinematographers even ones that do not have big budgets should have acess to a cinematic tool that gives them 100% control over exposure and sensor gain and records 100% of the sensor data so they can take best advantage of what they are recording without losses, losses can be decided on later and reversed if you record everything the sensor can put out.
The more people that have access to a manually controlled camera in a price range that is affordable, the more younder filmmakers there will be that will have the skills of a cinematographer into the future, they may make some mistakes, but so did people shooting film, and learning from mistakes gives you the ability to control the light to bring the vision to the screen. Replacing everything with CGI and running the camera on automatic under bad lighting do not make better movies just cheeper ones.
==
I working on a Linux Ubuntu port of my de-Bayer program, that can maybe reduce the cost of the render farm for final grading to 48bpp DPX frames. That issue is more important for 4K uncompressed DI than 2K because you have 4K the data and time involved and so would need 4x the number of render computers, using the free OS Ubuntu on junk computers can reduce the cost of setting up a render farm without loss of image quality.
Hopefully someday DCP will be uncompressed as well so people who sit in the front row can see what the sensor captured as well...
Last edited:
