A7s recording options

squig

squig

Veteran
I've been trying to decide between a Shogun, an Odyssey7Q, and a Pix 220i and after much deliberation I decided to go with a secondhand Sound Devices Pix 220. Here's why:

1) ProRes 4444 recording: it's all about the compression ratio. I really don't see the point in recording ProRes 422HQ 4k and then downscaling to 2k when I can just record ProRes 4444 1080p and do a 2k upscale.
2) Professional 702 level 4 channel synced sound recording (2ch over HDMI + 2ch over balanced XLRs)
3) I got it really cheap :grin:

I spoke to Mitch from Convergent and he said they are planning a ProRes 4444 upgrade for the Odyssey7Q, but unfortunately it doesn't have the pro audio capabilities of the Pix 220. The Shogun has an XLR breakout box but it's not known if it will be as quiet as a Sound Devices unit and there's no word from Atomos about ProRes 4444 recording. Ideally Sound Devices will bring out a 7" Pix recorder with LUTs and 4k ProRes 4444 recording.
 
I don't think your point 1) holds - 4K proresHQ is around double the data (read color info) of 1080p prores444 (and that's not taking into account that your 4:4:4 stream is really 8-bit 4:2:2) that means 2 times less compression. But 2) and 3) are probably good reasons to get a Pix anyway.
 
Last edited:
Also regarding 1: if you record 444 1080p, you do so with a 422 signal (if the HDMI-out is truly 422). If you record 4K, you have all the information you need for 444 1080p.

In any case, all these theoretical differences won't matter once we have the cameras and can test with them. Until then, it's just speculation...
(I plan on recording internally only, so I won't be able to help with this)
 
I did consider the 4k to 1080p 10bit conversion but in practice it doesn't sound like it's doing much to improve IQ: http://www.dvxuser.com/V6/showthrea...-10bit&p=1986424885&viewfull=1#post1986424885

4k Prores 422HQ is the same bit-rate per pixel as Prores 422HQ 1080p. The Prores 4444 bit-rate is 50% higher and I see a significant difference recording 8bit material with less artefacts particularly in the blacks.

It's going to be interesting to see how the internal 420 footage holds up against 422 prores 4444. The 50Mbit/s long-GOP codec should be very efficient. The Pix 220 doesn't do 1080p/60p or 720p/120p so I will have to rely on the internal codec for that.
 
I wasn't talking about the 10-bit thing, but about the 444 thing. In theory you can get 8-bit 444 with 10-bit in the luminance channel. But I've been fighting with the math to see how useful the 10-bit thing would actually be for the cases where we need it most, and wasn't sure it would help a lot, my final guess that it would be somewhat better than 8-bit but not as good as 10-bit, but I left it there just because my head was starting to hurt :p Looking at this very good example, it's pretty clear I should have just played with some footage to see what I get (it's not difficult, just get 8-bit 420 1080p footage and convert it to 960x540, and see if the gradation has improved).
 
I think you're going to be very happy with the internal codec. The Pix 220 makes more sense for me because I'll be shooting almost every day for a couple of months and it will be great to just open the ProRes 4444 files in an editor with 4 channels of synced sound then just cut, grade, and export to 2k.
 
Squig, you will not be disappointed by the PIX 220. It is a totally reliable piece of equipment, the quality of the sound is great and when you finished shooting you just plug in the caddy to you pc to copy, view or edit the files. I have been using it with a Nikon D7100 for a three weeks shoot and never had a problem with it.
 
Andrea189 said:
Squig, you will not be disappointed by the PIX 220. It is a totally reliable piece of equipment, the quality of the sound is great and when you finished shooting you just plug in the caddy to you pc to copy, view or edit the files.
Click to expand...

Sound Devices gear is top notch, I've got a mixpre-d that will be outputting to the Pix 220. I think the Shogun & Odyssey7Q will be great for video but they won't match the Pix 220 for sound recording.
 
Samuel H said:
just get 8-bit 420 1080p footage and convert it to 960x540, and see if the gradation has improved).
Click to expand...

This is exactly what I did before, when I was trying to simulate the case. I used some Canon DSLR sourced 4:2:0 footage - a heavily defocused colorful image with an abundance of color gradients in it. Downscaled in Resolve exactly two times (to 960x540) using the Smooth scaling filter. Resolve works internally in 32 bits, fyi. Then a single frame exported as 16 bit tiff (essentially, 10-bit luma, 8 bit chroma 4:4:4 image) and then the same frame as 8-bit 4:2:0 jpeg at a bitrate greatly exceeding any 4:2:0 video codecs in use. The first image simulated a 4k to 2k downscaled frame of 4:2:0 video as it will appear in the grading software if you work in a 2k timeline, the second simulated a 4:2:0 source video 2K frame as it would appear in grading. Then I imported both images in Photoshop and increased contrast a lot. There was no contest. Both were broken down to banding, but the "10-bit" image held much better.

Anyone can do a test like this, it takes 5 minutes, literally.
 
Last edited:
I love SD gear. But if I buy a recorder soon it will be the O7q, simply because it is pretty hot across many cameras.

C500, FS700. (A7s?, GH4k?)

I would consider buying a recorder with a view beyond the camera you are using today.

For example bringing in the 700 on HFR projects is very useful..

S
 
jcs said:
cpc- the difference in your examples appear to be due to compression artifacts and macroblocking related to DCT scale and noise vs virtual 10-bit luma. See my comments and images here: http://www.eoshd.com/comments/topic/6510-4k-8bit-to-2k-10bit-lets-get-to-the-bottom-of-this-please/
Click to expand...

I concur and I was never happy with the banding & macroblocking in ProRes 422HQ. ProRes 4444 looks a lot better due to the 50% higher bit-rate. Matched with the great looking 1080p out of the A7s I think the recorded image will look as good if not even better than 4k Prores 422HQ. Intra-frame encoding just isn't very efficient and I think the ProRes 422HQ bit-rate is just a little too low and the codec struggles a bit with problem areas like shadow detail. Sure I wish it was a true 10bit output but I don't need the extra resolution from 4k, I'm quite happy with the resolution out of the 5D raw. It's the extended dynamic range, low light ability, and no vertical fixed pattern noise that has me excited about the A7s.
 
Very nice test, cpc. It shows there's a very clear difference, even if it doesn't apply to people recording in even better codecs (like squig). The downsampling and encoding algorithms in Resolve are probably better than the ones you'll have in any camera, so I don't think the test is skewed against the 2K recording.
 
jcs said:
cpc- the difference in your examples appear to be due to compression artifacts and macroblocking related to DCT scale and noise vs virtual 10-bit luma. See my comments and images here: http://www.eoshd.com/comments/topic/6510-4k-8bit-to-2k-10bit-lets-get-to-the-bottom-of-this-please/
Click to expand...

It certainly is related to that. But not only.
There is always less tonal precision in the blacks because of noise and linearity. Once you push them up into grays territory in a physically correct way (i.e. trying to preserve relations between tones), it just isn't enough to sustain gradients anymore. Here the increased precision does actually help.

Also note that the "10-bit" image on the right is initially compressed (in "4K") with the exact same codec, just downscaled 2x in each direction. Which is pretty much the practical situation we are gonna be in. It is worth noting that the better codec you use (ideally uncompressed), the more true tonal precision you gain in the downscale. (That's mentioned in the text, I believe).

Thanks for the link, gonna check it in detail now.


@Squig: Well, I hope ProRes444 will hold. I just don't really believe it. s-log2 is quite flat. Some clever post-work can surely do wonders though.

@Samuel: Actually, the better codec you use, the more precision gain you get. :)
(Oh well, I just saw I repeat myself : )
 
Last edited:
You must log in or register to reply here.
Back
Top