What been happening up to now.
So much has already happened on this project so I expect this initial post to be edited hugely.
We just got off our 3rd or 4th monthly Skype conference call with the Atomic-VFX guys, where it was decided we should be documenting all our developments and discoveries as they happen. Resulting in this blog starting half-an-hour later. The goal is to help others considering similar Indie projects. Before I cover some of the points of this recent call, here is a quick overview of the project so far:
Spoon is an indie feature in development in South Africa. The team behind this feature contacted CineForm regarding using Prospect HD to complete the post work on this feature. The intention is to replicate some of the successes of Dust to Glory - an earlier feature onlined using Prospect. It turns out the Atomic-VFX guys are an adventurous bunch and planned to do much more with PHD by integrating into the product work-flow as a capture solution -- thereby bypassing camera compression for a much higher quality image suitable for effects-heavy work. Multiple camera are to be shot this way simultaneously. This is to be an indie project that does not shy away from the tricky stuff -- my understanding is that there are to be many effects shots, greenscreening, nights and rain shoots, all on a truly independent budget. To bring the costs down dramatically this feature will be shot with multiple JVC HD100/101 cameras at 1280x720 at 24p or 25p for the dramatic elements, overcracked to 50/60p for effects and slow-motion shots. They also intend to shoot through a mini-35 style lens adaptor to gain much better control of the depth of field. OK that is a tiny run-down of the project to date.
In more recent times the Atomic-VFX team has been testing a Prospect HD system (based on a dual AMD Opteron 275s) at live capture of 1280x720p60. As the current AJA Xena card only has HD-SDI inputs, the analog feed from the camera is sampled using an AJA HD10A unit to convert the 720p60 to HD-SDI. This was the first time a raw 720p60 feed was ever captured by PHD system -- so an issue did come up. Previously all 720p60 tests at CineForm came from Varicam DVCPRO-HD sources, which are somewhat heavily compressed and filtered (to 100Mb/s at 960x720 by the DVCPRO HD codec) which typical CCD sensor noise can't survive. So this raw signal from the JVC produced a bit-rate that exceeded our bit-rate cap, causing the noise characteristic to change (i.e. as expected there is more information in the raw JVC feed than a compressed DVCPRO-HD source.) Using CineForm compression this is supposed to be visually lossless so we needed to address this. Bit-rate tests indicated the image data demanded around 180-200Mb/s (a lot of data from our 3D wavelet tranform) and the signal was being clamped to rates around 150-160Mb/s. The simple solution is to lift the data-rate cap, which was previously put in place to allow slower workstations to capture without dropping frames -- this will be one of our next steps. Secondly, even with the current PHD release (1.1.0.32?) this issue will not occur for lower frames p50, p30, p25, or p24. 150Mb/s is plenty of data as the frame rate is reduced, as the cap is the same for all frame rates.
Capturing all the lower frame rates would be cool, but....it turns out the AJA HD10A sampler will only operate at 60p (or 59.94p), so 50p is out. Other supported frame rates over HD-SDI are 30p, 25p and 24p are pull-down versions of 50 or 60p, so 25p is out. Currently 24p and 30p works yet the data is captured as 60p, and the pulldown is not removed (so we get duplicate frames encoded.) Clearly the removal of the pull-down is now the highest priority as much of the production will be captured using the 24p mode on the JVC camera.
Other capture related issues revolve around the new Wafian ingest device. Atomic-VFX hopes to use these devices on this project so Wafian is working hard to meet that goal. Jeff, I hope you can expand on this here.
That's it for my first post.
David Newman
CTO, CineForm
We just got off our 3rd or 4th monthly Skype conference call with the Atomic-VFX guys, where it was decided we should be documenting all our developments and discoveries as they happen. Resulting in this blog starting half-an-hour later. The goal is to help others considering similar Indie projects. Before I cover some of the points of this recent call, here is a quick overview of the project so far:
Spoon is an indie feature in development in South Africa. The team behind this feature contacted CineForm regarding using Prospect HD to complete the post work on this feature. The intention is to replicate some of the successes of Dust to Glory - an earlier feature onlined using Prospect. It turns out the Atomic-VFX guys are an adventurous bunch and planned to do much more with PHD by integrating into the product work-flow as a capture solution -- thereby bypassing camera compression for a much higher quality image suitable for effects-heavy work. Multiple camera are to be shot this way simultaneously. This is to be an indie project that does not shy away from the tricky stuff -- my understanding is that there are to be many effects shots, greenscreening, nights and rain shoots, all on a truly independent budget. To bring the costs down dramatically this feature will be shot with multiple JVC HD100/101 cameras at 1280x720 at 24p or 25p for the dramatic elements, overcracked to 50/60p for effects and slow-motion shots. They also intend to shoot through a mini-35 style lens adaptor to gain much better control of the depth of field. OK that is a tiny run-down of the project to date.
In more recent times the Atomic-VFX team has been testing a Prospect HD system (based on a dual AMD Opteron 275s) at live capture of 1280x720p60. As the current AJA Xena card only has HD-SDI inputs, the analog feed from the camera is sampled using an AJA HD10A unit to convert the 720p60 to HD-SDI. This was the first time a raw 720p60 feed was ever captured by PHD system -- so an issue did come up. Previously all 720p60 tests at CineForm came from Varicam DVCPRO-HD sources, which are somewhat heavily compressed and filtered (to 100Mb/s at 960x720 by the DVCPRO HD codec) which typical CCD sensor noise can't survive. So this raw signal from the JVC produced a bit-rate that exceeded our bit-rate cap, causing the noise characteristic to change (i.e. as expected there is more information in the raw JVC feed than a compressed DVCPRO-HD source.) Using CineForm compression this is supposed to be visually lossless so we needed to address this. Bit-rate tests indicated the image data demanded around 180-200Mb/s (a lot of data from our 3D wavelet tranform) and the signal was being clamped to rates around 150-160Mb/s. The simple solution is to lift the data-rate cap, which was previously put in place to allow slower workstations to capture without dropping frames -- this will be one of our next steps. Secondly, even with the current PHD release (1.1.0.32?) this issue will not occur for lower frames p50, p30, p25, or p24. 150Mb/s is plenty of data as the frame rate is reduced, as the cap is the same for all frame rates.
Capturing all the lower frame rates would be cool, but....it turns out the AJA HD10A sampler will only operate at 60p (or 59.94p), so 50p is out. Other supported frame rates over HD-SDI are 30p, 25p and 24p are pull-down versions of 50 or 60p, so 25p is out. Currently 24p and 30p works yet the data is captured as 60p, and the pulldown is not removed (so we get duplicate frames encoded.) Clearly the removal of the pull-down is now the highest priority as much of the production will be captured using the 24p mode on the JVC camera.
Other capture related issues revolve around the new Wafian ingest device. Atomic-VFX hopes to use these devices on this project so Wafian is working hard to meet that goal. Jeff, I hope you can expand on this here.
That's it for my first post.
David Newman
CTO, CineForm