10bit 4:2:2 (pocket) vs 8bit 4:2:0 (5D3) Not Much Difference

For my review of the pocket camera I wanted to explore the difference between 10bit 4:2:2 (pocket) vs 8bit 4:2:0 (Canon 5D3). I have never used a camera with ProRes 10bit 4:2:2 before.

I underexposed the image for both the 5D3 and the Pocket exactly two stops and then tried to bring it back up in post using resolve.

How come I am not noticing much of a difference?

Watch my video of the test:

Try it with a not so well balanced scene. Something that well lit will be of marginal difference; but it's when you start getting into higher dynamic range stuff that you run into issues. A better test would probably be this:

Same scene, shot normal, then over expose in 1 stop increments until 4 stops over, then go down the same way to 4 stops under, then correct both back to normal.

@Adrian wow I thought 2 stops was a lot but I will try testing up to 4 stops next if you think that will be a good test.

Adrian is right, more DR would help with understanding where the ceiling is on both camera's footage.
Also, there could be other reasons why you aren't seeing the differences being larger. First some math: 8 bit color yields 16,777,216 possible colors. 10 bit color yields over 1 Billion possible colors. And for exponential reference, 12 bit is over 68 Billion possible colors.
Secondly, there is inherently more info in a 10 bit color space than an 8 bit. I think it goes without saying that the compression happening in camera in the BMPCC to Prores 422 HQ is leagues better than the 5DMKIII to H.264.
You don't list your reference equipment, so right there could be one problem. Your computer monitor is likely an 8 bit monitor. To really see the differences you need a broadcast reference monitor widely used in professional editing and grading bays. These monitors are usually 10 bit, or some even at 12 bit.
Another thing to keep in mind, is that you are compressing the video so much to be streamed that asking people to see what you are talking about via an internet streamed video is not the wisest way to start a convo. You should include the link for people to download the video at full resolution. With that said though, I do see a slight difference in the streaming video. The BMPCC skin tones is "rounder" or "more robust" in color than the 5DMKIII. In order words, I see slightly more color depth. I will safely bet you'd see more with a proper ref monitor.

You're comparing a $1K camera to a $3.5K camera (list prices). I don't think this addresses all of your question, but it's a significant consideration. I would expect a camera that costs several times as much as a BMPCC pocket cam to produce video that's "close to" or better than the BMPCC, at least in certain respects.

You're also comparing Apple's "ProRes 422 HQ" to Canon's implementation of "MPEG-4 AVC / H.264". I don't know anyone who would argue that ProRes 422 HQ isn't far, far more capable than MPEG-4 AVC / H.264. So, most of the difference/similarity shown in your test (or any comparison) with these 2 cams will have at least as much to do with the cam's hardware capability (sensor, internal processing, etc.) than their recording formats.

Finally, what you see on your computer is of course different than what we see in a compressed-for-the-web video on Vimeo. Web compression tends to equalize differences among cameras.

So, if the point of the comparison is to show there's not much difference between the video produced by a $1K camera and a $3.5K camera when viewed on Vimeo, then your test is a success.

And, in purely economic terms, the BMPCC (ProRes HQ) clearly wins: It costs a fraction of what a 5DM3 costs, yet produces video that's at least as good.

Once the BMPCC is capable of shooting RAW, I'd expect its price/performance ratio to improve further still.

-

Try with a scene with a lot of detail and a lot of different color-detail.

I have tried to put it against a D800, and while luma-detail is only a bit below, color-detail is way better on the BMPCC.
Also, it is much easier to grade, and can be pushed further without looking wierd.

At least, these are my findings.

You're comparing a $1K camera to a $3.5K camera (list prices). I don't think this addresses all of your question, but it's a significant consideration. I would expect a camera that costs several times as much as a BMPCC pocket cam to produce video that's "close to" or better than the BMPCC, at least in certain respects.

You're also comparing Apple's "ProRes 422 HQ" to Canon's implementation of "MPEG-4 AVC / H.264". I don't know anyone who would argue that ProRes 422 HQ isn't far, far more capable than MPEG-4 AVC / H.264. So, most of the difference/similarity shown in your test (or any comparison) with these 2 cams will have at least as much to do with the cam's hardware capability (sensor, internal processing, etc.) than their recording formats.

Finally, what you see on your computer is of course different than what we see in a compressed-for-the-web video on Vimeo. Web compression tends to equalize differences among cameras.

So, if the point of the comparison is to show there's not much difference between the video produced by a $1K camera and a $3.5K camera when viewed on Vimeo, then your test is a success.

And, in purely economic terms, the BMPCC (ProRes HQ) clearly wins: It costs a fraction of what a 5DM3 costs, yet produces video that's at least as good.

Once the BMPCC is capable of shooting RAW, I'd expect its price/performance ratio to improve further still.

-

Hey Peter I always see you making great contributions to this forum but It seems like you're comparing the prices, he's comparing specs. The price of these cameras are fairly irrelevant, which is probably why he didn't mention it anywhere.

Meaning if he were to use ProRes from the original BMCC, which is closer in price (even though its $1000 cheaper), the specs are what would still be relevant. That's is also considering BMPCC is basically a miniture BMCC... Just my 2¢.

Exactly, price hasn't got too much to do with performance on the whole. The main issue with the test is just the very low contrast, low detail, and low saturation of the original reference scene. There just isn't much too it. I could probably throw up any HD Camera and get similar results because the scene is just not that difficult to deal with.

One of the best examples of scenes which kill cameras are woods. Trees with leaves and bright sun and dark shadows. Even the Red had problems with such scenes when it was first released with it's "raw," breaking up in marcoblocks. It is here you will really start to see the differences in bit-rate come into play (such as was observed on the GH2 when hacked -v- unhacked for example).
On this level, I think the MKII it 50mbps, whereas ProResHQ can go up to something like 220mbps, that's substantially more data to deal with which should result in less macroblocking.

Another place to look is in gradients of color and contrast. Here is where 8 bit will start to break up and band which shouldn't be as prevalent on a 10 bit system-- though looking @ it on a computer display may look very similar because of the bottle neck of the displays quality.

On the whole, what I'm saying, is that in controlled situations the 5D and the BMs should be very similar-- as should most camera systems. But, it is in the real world outside of the studio where they start to show their stripes. Any camera can produce pleasing interview footage, well lit, and pretty properly exposed. But as you start to put them into more and more real world situations you start to encounter their limitations.

Seems like a lot of noise the 5D shadows.

Wow, lots of great stuff, I will redo my tests with the suggestions you have given.

Thanks everyone for sharing your knowledge, I have a ton to learn!

If you like I can post the retest here.

ps. I have a Samsung Syncmaster 244t monitor which is an 8-bit panel, I hope to get a new one soon!

I think this camera is going to take some time to work out fully. Certainly when it comes to this versus that conclusions.
I've been waiting about three weeks for the Bloom full review and final conclusions video. In that time, going by his tweets, he's been using the 'pocket' every day, in a variety of challenging circumstances, and alongside other gear he knows pretty well.
I'm assuming my first few weeks with this camera are going to yield fairly poor results.

Dave,
Your humility and willingness to listen and learn is generating some great threads. Keep up the good work.

@Chrisso, some days I feel like I am getting the hang of it, but most days I feel like I have a long way to go to get consistently good images out of the BMPCC.

Wow, lots of great stuff, I will redo my tests with the suggestions you have given.

Thanks everyone for sharing your knowledge, I have a ton to learn!

If you like I can post the retest here.

ps. I have a Samsung Syncmaster 244t monitor which is an 8-bit panel, I hope to get a new one soon!

When dealing with video, no computer monitor is adequate and that really isn't the workflow. You need a Decklink device to output to a properly calibrated SDI monitor. It is also possible to take output from your Decklink card to the HDLink and drive the HP Dreamcolor. But a computer monitor is an 8-bit panel that will not display in proper Rec709 color space.

Ned Soltz

ps. I have a Samsung Syncmaster 244t monitor which is an 8-bit panel, I hope to get a new one soon!

This is it.
Watch it on a Broadcast monitor or a large screen, say a cinema, and you will notice the difference.
Also, if doing effects or green screen you will notice.
As an exercise, view the footage at 400% of original scale and see the difference, analogous to watching on a large screen close up.

If you are using an Apple computer, the inbuilt graphics output is is 8 bit only, limited by the OS.

There is more to it than the bit depth. The computer display is in a different color space than a reference monitor fed by a capture or output card. You can then properly calibrate that monitor with bars and base all of your grading references from there.

It is great to see the learning strides that the OP is making but proper video display is crucial.

And to pick up on another theme, it is one thing to learn the tools at hand. That's the easy part. Developing the colorist's eye and skills is yet another. I can work with the tools and make things look better but wouldn't even begin to consider myself in the class of a professional colorist.

The simple fact here is that there is a world of difference between 8 bit H.264 and ProRes HQ

The big advantage for me to not using 4:2:0 (which I have done in the past) is the ability to put in your own grading style.

A lot of movies now have that annoying orange teal grade which you can force on your footage, and with BMCC ProRes too, not just RAW. So substantially changing the colours in the scene. Try doing that with 4:2:0 with H264 compression, there's just no way that's happening.

Jules

P.S. Who's to blame for this orange teal look anyway?

The big advantage for me to not using 4:2:0 (which I have done in the past) is the ability to put in your own grading style.

A lot of movies now have that annoying orange teal grade which you can force on your footage, and with BMCC ProRes too, not just RAW. So substantially changing the colours in the scene. Try doing that with 4:2:0 with H264 compression, there's just no way that's happening.

Jules

P.S. Who's to blame for this orange teal look anyway?

Colour science and skin tones. seriously. Go to adobe kuler choose complementary and find a skin tone and you'll see those orange teal colours.

The only issue I have with your test is that your benchmark is catering to the 5d.

The 5d is still the low light king and always will be. However the Pocket coming from the BMCC family of sensors needs light. It's counterproductive to the way you expose for this camera.

Zerbas to 100 and half a step down. How would the two compare then? What about when RAW arrives?

The only issue I have with your test is that your benchmark is catering to the 5d.

The 5d is still the low light king and always will be.

I would call you up on that. It maybe if you are shooting in a cupboard or something. we shot couple of weeks ago at night, outside in the street with street lights, in and outside a car with the BMCC using ProRes and 5D MkII.

Certainly the 5D MkII footage was brighter. I loaded the BMCC ProRes into Resolve, on the scopes it looked collapsed down the bottom. Did an initial grade so got it to spread out in the scopes. Used Resolve 10 beta with the new NR options. Still had a lot of latitude to bring the skin tones out and do some hue changes etc. The result of the BMCC footage looked much sharper, much more filmic and generally better consistency. The 5D looked a bit videoish, soft, but it was bright.

Anyway not a scientific test. But saying 5D is king in low light is not my experience when it comes to which has the better footage after you grade it.

Jules

For my review of the pocket camera I wanted to explore the difference between 10bit 4:2:2 (pocket) vs 8bit 4:2:0 (Canon 5D3). I have never used a camera with ProRes 10bit 4:2:2 before.

I underexposed the image for both the 5D3 and the Pocket exactly two stops and then tried to bring it back up in post using resolve.

How come I am not noticing much of a difference?

Watch my video of the test:

try keying anything. you'll see the difference immediately when you look at the matte. You can also view what each channel looks like, and compare black and white samples.

The only issue I have with your test is that your benchmark is catering to the 5d.

The 5d is still the low light king and always will be.

I would call you up on that. It maybe if you are shooting in a cupboard or something. we shot couple of weeks ago at night, outside in the street with street lights, in and outside a car with the BMCC using ProRes and 5D MkII.

Certainly the 5D MkII footage was brighter. I loaded the BMCC ProRes into Resolve, on the scopes it looked collapsed down the bottom. Did an initial grade so got it to spread out in the scopes. Used Resolve 10 beta with the new NR options. Still had a lot of latitude to bring the skin tones out and do some hue changes etc. The result of the BMCC footage looked much sharper, much more filmic and generally better consistency. The 5D looked a bit videoish, soft, but it was bright.

Anyway not a scientific test. But saying 5D is king in low light is not my experience when it comes to which has the better footage after you grade it.

Jules

No doubt the BMCC resolves a greater image and has greater flexibility in post (which is the point of shooting in greater bit depth/chroma sub-sampling) but if there is a claim the 5d3 could make over the BMPCC line is that has a far greater usable gain range and does well in low light.

But I suppose that brings me to the initial qualms of the test. people inquiring about the OP's monitor have it right. Even with JB's sample footage, the image was tremendously malleable in post. Even before the grade, it was quite evident this was far superior to any 420 DSLR image.