Who will break the interenal 10 bit hybrid barrier?

[Guest Ebrahim Saadawi]

Reworking 1 still image from a raw file to 8 and 10 bit files.... and compressing video to 8 bit and 10 bit are two very different things.

This is why jpgs on a DSLR are far better than a framegrab. There is time to properly map the 14 bits to 8 bits.

I disagree. I believe reworking a raw 14bit still to 10bit and 8bit and comparing is the most proper way to compare 8 vs 10bit scientifically. 

JPEGs from SLRs look worse than frame grabs, even though they're both 8bit, is because the Jpegs do an elegant downscale, from the full sensor (no binning/skipping), are 4:2:2 chroma, much higher file size (less compression), and the fact that they use a mechanical shutter, not a rolling shutter. Frame grabs look bad because they're H.264, with line skipping/pixel binning, NR, heavy compression blocking, 4:2:0 sampling, and rolling shutter artefacts. Not 8bit. 

On the 1DC, the 4K files look identical to the Medium size JPEGs. Both 8bit 4:2:2. Just rolling shutter artefacts in fast pans. 

My point is I want manufacturers to get all of these other elements right first rather push 10bit upon us. 

Answering the title: My money is on the Panasonic Successor to the GH4/AF100. at the last quarter of 2016 (well BM did it first but I mean the next one). Sony, Canon, not now, not even close. Panasonic showed interest in 10bit with the GH4 output, and they wil NEED new headline features after such a full featured predecessor. Internal IS, 10bit, Raw out, etc. 

[Jimmy]

I would argue that 10 bit with decent compression is still better than 8 bit uncompressed.

I'll see how your test idea stacks up though with a heavy grade on a raw file from my a7s when i get chance. Maybe you are right.

[jgharding]

I'd like 10-bit, but have been waiting so long and have gotten so used to 8-bit i keep forgetting about it.

Not all 8-bit footage is equal mind you, 8-bit C100 is great compared to some 8-bit stuff.

I wish one of the pro-sumer companies would just do a Red style compressed raw and be done with it.

[AaronChicago]

C100 8 bit is pretty incredible for 24/Mbps. I've tried a A/B tests with the Ninja Star Pro Res and had to do a 400% zoom with a heavy grade to notice ANY difference. I'd say 90% of shots internal is good enough on that camera.

[Oliver Daniel]

There is definitely a difference between 8 bit and 10 bit. It just depends how well the camera utilises the feature with all that tech inbetween...(and other boring obvious statements). 

No need to get angry about it though. Why be emotional about bit rates? What a boring thing to be emotional about. The world is out there, somewhere, isn't it? 

[Mattias Burling]

There is definitely a difference between 8 bit and 10 bit. It just depends how well the camera utilises the feature with all that tech inbetween...(and other boring obvious statements). 

No need to get angry about it though. Why be emotional about bit rates? What a boring thing to be emotional about. The world is out there, somewhere, isn't it? 

I recently learned that when a person is angry and worked up brain cells are physically deteriorating. 

Which means that when you're mad, you are literally self destructing. 

Worth it over bit rate?  

[Nikkor]

Someone with a bunch of cameras could shoot a scene with 8bit and 10 bit log/flat profiles and en ask photoshop to count the colors, I can imagine the numbers will correlate with the feeled quality of these images.

[Xavier Plagaro Mussard]

The 8-10 bit is just one of the many variables! 

[Guest Ebrahim Saadawi]

I would argue that 10 bit with decent compression is still better than 8 bit uncompressed.

I'll see how your test idea stacks up though with a heavy grade on a raw file from my a7s when i get chance. Maybe you are right.

I'll try it too. Let's really see the difference scientifically. 

What codec will you use to downsample? 

We need a codec that has the exact same bit rate, chroma subsampling, compression method, just with an 8bit and 10bit versions. 

And render the 14/12bit raw image to each one. 

I'll see if DNxHD has similar 8bit and 10bit codecs.

If you're on mac, I think Customizing ProRes is the best option. Say put both at 350mbps, 4:2:2, one 8bit and one 10bit. I think that's possible. 

[Don Kotlos]

I disagree. I believe reworking a raw 14bit still to 10bit and 8bit and comparing is the most proper way to compare 8 vs 10bit scientifically. 

Ebrahim, if you do that then the difference will be minimal indeed. The limitation of the 8bit/channel is when you are using a log curve as I tried to argue here. The reason is that you have to allocate a very small number of colors to each stop of the dynamic range. The reason that canon 8bit is great is because they are using a less aggressive log curve, that maps many more steps on the stops around mid-tones where the skin mostly lies. See here: http://www.xdcam-user.com/2011/12/canon-c-log-on-the-c300-compared-to-s-log/

The test is to have a camera with an aggressive log such as the s-log2 of the Sony and a 10 bit output. Record at 10bit and convert the ungraded version to an 8bit version too. Then try to grade both. 

[sunyata]

Someone with a bunch of cameras could shoot a scene with 8bit and 10 bit log/flat profiles and en ask photoshop to count the colors, I can imagine the numbers will correlate with the feeled quality of these images.

to calculate number of possible colors per bit depth, the formula is 2^n where n=bit depth. 1bit would be 2 colors, 4bit would be 16 and so on.  photoshop will call 8-bit "thousands of colors" but that's like marketing-speak, they're including all intensity levels, so 256*256, not only chromaticity values. 8-bit will always be 256 colors and 10-bit will be 1024.

[Nikkor]

There is an option that let you count the number of singular colors present in a frame, depending on the processing the camera does there can be a lot of them or just a few.

[Don Kotlos]

to calculate number of possible colors per bit depth, the formula is 2^n where n=bit depth. 1bit would be 2 colors, 4bit would be 16 and so on.  photoshop will call 8-bit "thousands of colors" but that's like marketing-speak, they're including all intensity levels, so 256*256, not only chromaticity values. 8-bit will always be 256 colors and 10-bit will be 1024.

?

8bits/channel   --> 24bits/ image -->     16777216 colors. 

10bits/channel --> 30bits/ image --> 1073741824 colors. 

This is theoretically what can be recorded. 

Practically,as I said in a previous thread, when recording skin with log 8bit image you will mostly get it within 1-2 stops of the dynamic range which reduces the color space dramatically. 

[AaronChicago]

[sunyata]

i was just referring to chromaticity. https://en.wikipedia.org/wiki/8-bit_color, or chroma separate from luma per channel. from above link on 8-bit color "The maximum number of colors that can be displayed at any one time is 256."

i checked after effects which i don't use that much anymore, sorry, they do refer to 8-bit as millions of colors (for some reason i thought they called it thousands), my bad, big brain fart (referring to me).

 

[Don Kotlos]

you're thinking of bits per pixel? which does not determine number of colors. https://en.wikipedia.org/wiki/8-bit_color color as defined by chromaticity, or chroma separate from luma. from above link on 8-bit color "The maximum number of colors that can be displayed at any one time is 256."

In the link you provided the difference in colors between 8bit and 10bit video is between "True color (24-bit)" and "Deep color (30-bit)". 

8/10bit video means per channel per pixel per frame.  

The two chromaticity channels (luminance independent) are useful when trying to describe displays since the luminance can be changed very easily and when compressing the video signal. 

[Guest Ebrahim Saadawi]

Ebrahim, if you do that then the difference will be minimal indeed. The limitation of the 8bit/channel is when you are using a log curve as I tried to argue here. The reason is that you have to allocate a very small number of colors to each stop of the dynamic range.

Okay I'll do the 14bit to 10bit vs 8bit with both Log and normal gamma. Perhaps it's Log where 8bit will show its face. 

The reason I believe testing from a single still image is the best way is because we remove all the variables, every single one. Just purely testing 8bit vs 10bit effect on an image. But if we record 10bit with one camera and 8bit with another, well massive variables enter here. Even if it's the same camera, the 10bit and 8bit in an FS700 have a ton of variables, and even GH4 10bit vs 8bit will have different codecs and different NR, bitrate, so skipping all that we begin with a single, simple rich source. 

and supersampling is good to ensure we're testing perfectly full 10 and 8 bit, don't want claims that 10bit on a GH4 is really 9bit and so on (Which I have heard).

I am still searching for a codec that has the same characteristics in 8bit and 10bit versions. Any thoughts? (even a still photo compression format, or a codec). Windows Doesn't create ProRes sadly. And DNxHD has vastly different bitrates for 8bit and 10bit, 10bit being MUCH higher, so will not cut it for the test. 

If anyone willing to do the test please share. Just any Raw image file you have on you drives and a way to compress it to 10 and 8 bit without variables. 

 

[Don Kotlos]

Okay I'll do the 14bit to 10bit vs 8bit with both Log and normal gamma. Perhaps it's Log where 8bit will show its face. 

The reason I believe testing from a single still image is the best way is because we remove all the variables, every single one. Just purely testing 8bit vs 10bit effect on an image. But if we record 10bit with one camera and 8bit with another, well massive variables enter here. Even if it's the same camera, the 10bit and 8bit in an FS700 have a ton of variables, and even GH4 10bit vs 8bit will have different codecs and different NR, bitrate, so skipping all that we begin with a single, simple rich source. 

and supersampling is good to ensure we're testing perfectly full 10 and 8 bit, don't want claims that 10bit on a GH4 is really 9bit and so on (Which I have heard).

I am still searching for a codec that has the same characteristics in 8bit and 10bit versions. Any thoughts? (even a still photo compression format, or a codec). Windows Doesn't create ProRes sadly. And DNxHD has vastly different bitrates for 8bit and 10bit, 10bit being MUCH higher, so will not cut it for the test. 

If anyone willing to do the test please share. Just any Raw image file you have on you drives and a way to compress it to 10 and 8 bit without variables. 

 

You record 10 bit with one camera. Then can you convert that to an 8 bit video very easily. 

[TheRenaissanceMan]

As long as we're testing, I'd also love to see someone with an external recorder take a 10-bit capable camera like the GH4 or BMPCC and record the uncompressed out in both 8 bit and 10 bit. This theoretical stuff with RAW frames is interesting, but those conditions are where I see big differences in rendering. 

For example, this video was pretty eye-opening to me. It's a LUT that attempts to imitate Dragon color on the GH4, and you can clearly see in the examples how much more natural the image is from the 10-bit HDMI after the Dragon LUT and standard grading are applied. It's a drastic difference. In many cases, the color palette is completely different between the two, because the 10-bit captures shades the 8-bit can't even see. 

https://vimeo.com/101350338

This gap in aesthetic is night and day different to the gap between compressed and uncompressed 8-bit from the Nikon D5300, for example. 

And quick question. If your main argument against 10-bit is that the larger file sizes aren't worth it, how can you argue for the 1DC shooting 500 mbps files? Even Prores HQ, a 10-bit 4:2:2 codec, is only 220 mbps, and most people could shoot LT without noticing any degradation. Panasonic's 10-bit 4:2:2 AVC-Ultra codec is even more efficient at half the size in 1080p and almost a third the size in 4K. How does it look? You tell me. 

https://vimeo.com/105522587

[Fritz Pierre]

The difference between 8 & 10 bit would be much more apparent....using the camera all day...lit set...$150,000 a day crew...lots of actors...a narrative and scenes that actually have to cut together...that would be a real test...I'm a bit off topic Renaisance Man:):)...