BMPCC: what actual latitude in post?

Hi there,

Like many of us, I've pre-ordered a BMPCC and was wondering what actual latitude I shall expect from it in post in ProRes?
Using the street sample footage, I've tried to recover shadow details but found the image to quite easily collapse when pushing just a bit more than what I usually do with my hacked GH2's footage.
I am quite new to everything post/color correction, so I would appreciate to have some specialists' view on the topic.
Thanks!

As someone who has worked on legit hollywood films in post, I actually find even the 14 bit raw that you can get with a hacked canon DSLR to be a bit shallow as I'm used to working with 16 bit stuff you get out of REDs and Alexas. I'll be honest and admit that when I read that the raw footage that BMD cameras output is only 12 bit I was disappointed. So, with only two extra bits of depth to work with, you will see some improvement in latitude over the 8 bit that most DSLRs put out, it still isn't that much better.


That being said, the fact that you get prores out of the BMPCC is still lightyears ahead of everything else even remotely within that price range(barring the 5dIII with magic lantern). The noise that you get with the prores is vastly better than that horrible blocky mess that you get with h.264(looks more like film grain than compression artifacts). So much that I would love if A1ex would put out a version of ML which records videos as jpeg sequences in stead of just raw; even with the loss of color bit depth, so long as it gets the video resolution closer to 1080p on my T3i. Though, I gather that such a thing is not possible.

As someone who has worked on legit hollywood films in post, I actually find even the 14 bit raw that you can get with a hacked canon DSLR to be a bit shallow as I'm used to working with 16 bit stuff you get out of REDs and Alexas. I'll be honest and admit that when I read that the raw footage that BMD cameras output is only 12 bit I was disappointed.

The BMCC and Pocket are both greater than 16 bit from the sensor, 16 bit linear internally before converting to 12 bit LOG to record to DNG. You'll notice the 12bit files "unpack" in Resolve (or PS) as 16 bit.

RED are secretive about what's happening internally, and tend to dodge that question when asked directly but one presumes the same things are happening, because sometimes they seem to say 16 Bit and sometimes they say 12 bit.

jb

The BMCC and Pocket are both greater than 16 bit from the sensor, 16 bit linear internally before converting to 12 bit LOG to record to DNG. You'll notice the 12bit files "unpack" in Resolve (or PS) as 16 bit.

That does explain one thing, which is that the BMCC's latitude far exceeded my expectations. I spent a week in Glacier National Park in July, and since I don't yet have IRNDs I didn't use them (they're on my list of stuff to acquire when budget allows). Due to the trip schedule, I ended up doing most of my photography during the middle of the day, and we were between five and six thousand feet most of the time with clear blue skies, which means crazy harsh light, harsh enough that you'd think twice about even using black and white negative film.

The only things that were blown out because they exceeded the range I could capture with my BMCC was snow in direct sunlight.

That does explain one thing, which is that the BMCC's latitude far exceeded my expectations.

Same here. During the first test I did with the BMCC, it was snowing in London. I haven't dealt with RAW footage from the RED or ALEXA cameras yet, but the dynamic range of the BMCC sure is impressive.

Before:


After:


DNG download: http://blog.christianschmeer.com/conten ... 1/snow.zip

As someone who has worked on legit hollywood films in post, I actually find even the 14 bit raw that you can get with a hacked canon DSLR to be a bit shallow as I'm used to working with 16 bit stuff you get out of REDs and Alexas. I'll be honest and admit that when I read that the raw footage that BMD cameras output is only 12 bit I was disappointed. So, with only two extra bits of depth to work with, you will see some improvement in latitude over the 8 bit that most DSLRs put out, it still isn't that much better.


That being said, the fact that you get prores out of the BMPCC is still lightyears ahead of everything else even remotely within that price range(barring the 5dIII with magic lantern). The noise that you get with the prores is vastly better than that horrible blocky mess that you get with h.264(looks more like film grain than compression artifacts). So much that I would love if A1ex would put out a version of ML which records videos as jpeg sequences in stead of just raw; even with the loss of color bit depth, so long as it gets the video resolution closer to 1080p on my T3i. Though, I gather that such a thing is not possible.

10 bits is a ProRes output. Raw is 12.
12 bit and even 10 bit ARE vastly better than 8. Even 9 bit would be vastly better. The improvement in sampling rate is not a "2 more" between 8 bit and 10 bit, it's 4 times better, for 12 bit it is 16 times better than 8 bit. It's an exponent operation.

What's you real name? This is a real names forum.

Thanks for your replies - very interesting.
That snowy London shot is impressive for sure!
Now, we don't really know what to expect from raw on the Pocket camera, so what's your hands-on experience with its ProRes files latitude?

Same here. During the first test I did with the BMCC, it was snowing in London. I haven't dealt with RAW footage from the RED or ALEXA cameras yet, but the dynamic range of the BMCC sure is impressive.

One of my mentors spent some time playing with some of the Glacier NP footage in Resolve. He took a shot that was so overexposed that the entire sky was a wash of white, and you could barely see the texture of the waves in the foreground. It took him about two minutes to grade it to the point where not only was the texture of the waves beautiful and detailed, but so were the mountains in the distance...

Of course the <bleeping> mac crapped out while he was doing that and as a result I wasn't able to save it, so I'll have to grade it again before I can share it. Sigh...

This is all great for raw users, but what about the Pocket camera's ProRes?

I have some TV stuff in prorez. Also really nice if you think what else you have for that price range
and how well it holds in grading.

Some people here i know are only shooting Prorez as their macs are dated laptops

I myself will go into production now with raw.

Look here for the prorez spots:

https://vimeo.com/iconstudioseu

Those last 4 are all prorez and graded in resolve.

Even the ProRes files have a good range and the highlights roll off really nicely.

Here is some test footage I shot for the Antler Post Film to REC.709 FCPX plugin:
http://www.adamroberts.net/blog/blackma ... tler-post/

To be more specific, here is what I manage to recover from the shadows with a hacked GH2:

Original (trying not to burn the highs):


Shadows lift:


Imagine I would have shot this on the Pocket camera, would I have been able to recover more details from the shadows? Would the highs have not clipped?

Even the ProRes files have a good range and the highlights roll off really nicely.

Here is some test footage I shot for the Antler Post Film to REC.709 FCPX plugin:
http://www.adamroberts.net/blog/blackma ... tler-post/

Beautiful!
Let me try to dig some more shadow details from that Pocket cam's footage and post my impressions.

I have some TV stuff in prorez. Also really nice if you think what else you have for that price range
and how well it holds in grading.

Some people here i know are only shooting Prorez as their macs are dated laptops

I myself will go into production now with raw.

Look here for the prorez spots:

https://vimeo.com/iconstudioseu

Those last 4 are all prorez and graded in resolve.

Excellent job! That white shirt would have been tricky with a lower DR DSLR for sure.

Try Adobe Premiere cc LUT using the new built in Lumetri plugin in ...
There is many great lut , you will be

By memory i'd say i got to take some 25% down the hi lights without any worries in prorez.

So it holds up pretty well. Zebra was set to 95% in cam and film mode gamma.

If i'd own dslr i would smash it right a way if not hacked

What dslr also misses is possibility to run without overheating.
It was really bad few years back with clients on the set when body just overheats after every 10 minutes.

In my mind even without uncompressed raw BMC and BMPC we got really huge deal in our hands.

With my limited understanding of color correction, I've tried to fiddle with JB's ProRes Pocket files.

Original exported from Premiere:


With a slight shadows lift and a little bit of highlights recovery (kind of) - more than that doesn't look right to me:


Trying to recover as much detail in the shadows as possible, it quickly becomes over-grainy and simply unusable:


Overall, as soon as I touch the original, I find the noise to be distracting.

[CaptainHook]

The BMCC and Pocketcam tend to work better the other way, rather than lifting shadows you get better results pulling down highlights (that haven't clipped obviously) than you would with a DSLR. If you want to push shadows, i've found it "better" (less noise) to shoot at 400 or 200ASA in Prores although you take a slight hit in DR.

If i'd own dslr i would smash it right a way if not hacked

What dslr also misses is possibility to run without overheating.
It was really bad few years back with clients on the set when body just overheats after every 10 minutes.

In my mind even without uncompressed raw BMC and BMPC we got really huge deal in our hands.

Well, no BM camera has an articulated screen, a viewfinder, can take stills, has a battery life longer than a DSLR, has proper vu-meters, decent audio, can edit/delete clips in-camera... so - yes - I'm also excited about the new cameras offer but I wouldn't smash my DSLR yet

The BMCC and Pocketcam tend to work better the other way, rather than lifting shadows you get better results pulling down highlights (that haven't clipped obviously) than you would with a DSLR. If you want to push shadows, i've found it "better" (less noise) to shoot at 400 or 200ASA in Prores although you take a slight hit in DR.

aka "expose for the shadows" rather than for highlights, right?
This is good to know but still, I guess in a high contrast situation, digging information in those shadows will be crucial and from that quick test with JB's files, I'm not that certain that it is really possible.
I know that raw would take the Pocket cam to another level, but as it's not yet implemented, I consciously don't take that aspect into account.

If you grade you have to use scopes to evaluate results.

Only eye balling will not cut it and would be far from precise.

If you grade you have to use scopes to evaluate results.

Only eye balling will not cut it and would be far from precise.

I always have RGB parade open when grading. Sometimes, everything looks fine on the scopes but the image is simply ugly. This is the case with this last shot I posted.

Look histogram and mind your black and middle tones.

I'm asking those questions because banding, macro blocking, little DR to play with, rolling shutter etc. are the things I'd like to forget when upgrading to a new camera.
I know both BMCC and BMPCC have a rolling shutter so, at least, I'd like to make sure latitude is really there, even in ProRes and I must say I've yet to be convinced from my own tweaking.

Has some of the color correction gurus here actually played with that same sequence from JB?
I'd be interested to see how much shadow details they have/would be able to recover from that pavement zone and hear their feedback.

How much are you expecting out of a one-thousand dollar camera, that's the real question. It's already doing things that cameras thirty times the price tag aren't...

The scene was captured within a range that protects the sky. So, of course most of the bottom end's going to be noisy. Thirteen stops isn't enough to capture that kind of scene entirely, some sacrifices have to be made. I'm guessing John foresaw the cries about limited DR in a blown out sky and decided not to clip it.

That's what I would have done.

The dark areas should be dark, leave them that way and you'll get an idea of where the image is supposed to be.

Or split the scene with multi nodes controlling separate parts of the image and combine them.

In resolve you can also do noise reduction to the shadow part if you have full license and cuda gpu.

How much are you expecting out of a one-thousand dollar camera, that's the real question. It's already doing things that cameras thirty times the price tag aren't...

Well, the Pocket cam is supposed to be a "true digital cinema camera" from BMD words and it's also supposed to have the same DR as the 2.5k model. So I'm not comparing it against its price tag (which is undoubtfully great), I'm trying to see what it can/can't actually do, beyond anything marketing etc.

The scene was captured within a range that protects the sky. So, of course most of the bottom end's going to be noisy. Thirteen stops isn't enough to capture that kind of scene entirely, some sacrifices have to be made. I'm guessing John foresaw the cries about limited DR in a blown out sky and decided not to clip it.

That's what I would have done.

Me too, obviously, even though some say that those cams can handle highlights better than shadows, so exposing for the shadows could have made more sense in this specific shot, apparently.

The dark areas should be dark, leave them that way and you'll get an idea of where the image is supposed to be.

Isn't it the beauty of great DR to recover information in post? Most of the BMPCC grades I've seen had their shadows crushed, which does not showcase how much information can be retrieved in this end of the spectrum, thus my questions.

I'm not trying to put this camera down, just trying to be realistic and a bit more specific than the average: "yeah, amazing, especially for the price!".

Isn't it the beauty of great DR to recover information in post? Most of the BMPCC grades I've seen had their shadows crushed, which does not showcase how much information can be retrieved in this end of the spectrum, thus my questions.

The beauty of extended DR is capturing more information than a camera with less DR, not to recover information in post. 13 stops on one camera isn't the same on another. Most of Epic's DR is in the bottom end, most of Alexa's the top end. Start with a strong base, finish with one ideally.

The beauty of less compression, bits, etc. is image fidelity and manipulation. This is still within the parameters of camera specs. You're trying to lift 10-bit ProRes Files which are a lot better than a ton of other cameras, but still isn't going to be magic.

If you want cleaner shadows you're going to sacrifice highlights. That's how it works on nearly every camera, anyway, some are better (a lot are better) in the low end, but very few are as strong in the top end. VERY few.

As far as price goes, it's still 1000.00, so it is pretty amazing for what it's doing -- it's not about putting the camera down, the images are there to review, not much is going to change from operating a traditional digital camera... the less light hitting the sensor the more noise, the less compression/less image manipulation in camera the more apparent noise becomes.

Thanks, that makes sense.
I'm looking forward to seeing what raw in the Pocket cam can do. In the meantime, I understand that ProRes is already better than - say - hacked GH2 footage. I was just hoping for more tweakability in post.
And, yes, the features packed in this $1k body still seems very interesting.

This camera doesn't conjur miracles, but it gets its job done pretty well. Like every other instrument of the trade you need to learn its strenghts and weaknesses, and how to exploit the first and avoid the second. Ten bits and ProRes compression is a fair balance between maximum IQ and handleable amounts of data, but we will only see the true potential of this sensor when they add compressed RAW to the firmware.

[CaptainHook]

aka "expose for the shadows" rather than for highlights, right?
This is good to know but still, I guess in a high contrast situation, digging information in those shadows will be crucial and from that quick test with JB's files, I'm not that certain that it is really possible.

Have you tried downloading my original Prores files from the Pocketcam? The daytime stuff was at 400ASA so maybe try lifting the shadows on those and see if they get too noisy? (i haven't tried myself) Link to files is in the description :

It's already doing things that cameras thirty times the price tag aren't...

Ive already had a good laugh, but am open to a few more.. Please, share some of these quantifiable "things"..

Hey, good question, Theodore.

Seeing as how it only does a handful of things to begin with, the easier to name are:

10- bit ProRes/DNxHD Onboard Recording (no peripherals needed)
DR 13 stops

Name more than three cameras at any price range, as a matter of fact, that can capture directly these highly requested 10-bit industry standard formats and has 13 stops, even on paper. I'll take paper specs.

We don't have to tack on doing it to SDXC cards (relatively cheap reliable media) or the future RAW update, since one I agree would be low-hanging fruit and the other isn't out just yet.

10 bits is a ProRes output. Raw is 12.
12 bit and even 10 bit ARE vastly better than 8. Even 9 bit would be vastly better. The improvement in sampling rate is not a "2 more" between 8 bit and 10 bit, it's 4 times better, for 12 bit it is 16 times better than 8 bit. It's an exponent operation.

What's you real name? This is a real names forum.

That may be, but 16 bits is still vastly more bit depth than 10 and 12. Whether even just 9-bit color depth is a considerable improvement over 8-bit is irrelevant. In the grand scheme of things, 10 bit is not that much of an improvement when one considers the existence of 16 or 32-bit footage. This is all subjective afterall, unless we can actually see what the OP means by "pushing just a bit more than hacked GH2 footage" we're all just making stabs in the dark. Maybe the OP has a different standard definition of a "bit more" of latitude in regards to color correction than you or I.

That may be, but 16 bits is still vastly more bit depth than 10 and 12. Whether even just 9-bit color depth is a considerable improvement over 8-bit is irrelevant. In the grand scheme of things, 10 bit is not that much of an improvement when one considers the existence of 16 or 32-bit footage.

You're not making any sense. In the grand scheme of things, going from the mush that dSLRs produce to 10-bit 4:2:2 ProRes is a big improvement, and going to 12-bit raw is huge.

The fact that there are cameras that can record 16-bit raw doesn't change that. It's true that if I had no budget constraints I'd go for an f55 or an Epic Dragon over a Black Magic, but since that's not the case, I'm using a Black Magic... and enjoying it. (So are the people I'm delivering my footage to. )

[CaptainHook]

You guys seem to keep overlooking a point John always tries to hammer home, the BMCC is 16bit raw and stores it as 12bit log, but when grading in Resolve or whatever it 'unpacks' as 16bit linear again.

That may be, but 16 bits is still vastly more bit depth than 10 and 12. Whether even just 9-bit color depth is a considerable improvement over 8-bit is irrelevant. In the grand scheme of things, 10 bit is not that much of an improvement when one considers the existence of 16 or 32-bit footage.

You're not making any sense. In the grand scheme of things, going from the mush that dSLRs produce to 10-bit 4:2:2 ProRes is a big improvement, and going to 12-bit raw is huge.

The fact that there are cameras that can record 16-bit raw doesn't change that. It's true that if I had no budget constraints I'd go for an f55 or an Epic Dragon over a Black Magic, but since that's not the case, I'm using a Black Magic... and enjoying it. (So are the people I'm delivering my footage to. )

The point is that the OP is talking about not having "much more" bit depth than with a hacked GH2, but without really giving a good frame of reference. Sure, he showed some before and after pictures, and mentioned some general terms in regard to what he did to the footage in question, but we don't know what methodology he used exactly. Did he go in on a channel-by-channel basis or did he just gang-adjust the whole thing? Did he jack up the contrast or the saturation? Did he use scopes or simply eyeball it? Does he know which channel has the most noise? Did he transcode the footage before or after applying the color correction? We don't know. Heck, unless we have established that the supposed problem that the OP is having is unique to the prores footage put out by the BMPCC, then he's not even asking the right question, as the issue is about the codec, not the camera.

The point is that saying something as subjective as "a bit more latitude" when talking about color depth can mean something very different if you've only ever worked on 8-bit h.264 footage than if you've spend most of your career working on footage from Alexas and REDs.

Five gallons is a lot when you're buying milk at the grocery store, it is virtually nothing when you are trying to fill a swimming pool.

RED are secretive about what's happening internally, and tend to dodge that question when asked directly but one presumes the same things are happening, because sometimes they seem to say 16 Bit and sometimes they say 12 bit.

jb

MX is 12bit, Dragon is 16bit

The BMCC and Pocketcam tend to work better the other way, rather than lifting shadows you get better results pulling down highlights (that haven't clipped obviously) than you would with a DSLR. If you want to push shadows, i've found it "better" (less noise) to shoot at 400 or 200ASA in Prores although you take a slight hit in DR.

Smaller sensor equals more noise. This detail is always overlooked. Will be interesting to see what the 4K camera will produce in terms of noise.

You guys seem to keep overlooking a point John always tries to hammer home, the BMCC is 16bit raw and stores it as 12bit log, but when grading in Resolve or whatever it 'unpacks' as 16bit linear again.

hear hear

The dynamic range or "latitude" of sensor in real life applications can be measured. Eg here:

http://www.dxomark.com/index.php/Camera ... ings/(type)/usecase_landscape

Nikon D800 reaches 14.4 Evs ("stops") and is #1 currently among all cameras tested by Dxo.
That's about the same real life latitude as that of Alexa (smaller sensor, but bigger pixels).
GH1 is 11.6Evs on that list. GH3 improves to 12.4 evs. Canon 5D m3 is pitiful 11.7 Evs. In case of photo cameras the Evs figures are for the whole sensor in photo mode. In film mode they must be worse, and get much worse when recorded with long GOP/low bit rate codecs like h264. Best to record in uncompressed 422 where possible.
All those sensor Evs latitude numbers are for raw. Codecs will decrease the performance. Anybody working with photos knows the difference between working with raw/psd files and the same pictures in jpeg. It can very well be compared to working in raw/dpx and Prores422 in movies. Prores444 would be like a very light jpeg (3:1). With prores and such, some parameters like white balance are baked-in and cannot be set anew in post like in case of raw.
BMPCC is advertised to have 13 stops of latitude. That presumably for the raw readout from the sensor.
In case of a mathematically lossy raw compression (similar to red 5:1 or so)* the flexibility in post won't be as good as with Alexa raw files, but possibly not far behind Red's. Prores422 (even HQ 220 kbit/sec) will give trouble in case of the green screen or Fx work.
*Any image compression above ca 2:1 must be mathematically lossy. For higher compression ratios manufacturers often use the phrase "visually lossless" but that's in the eye of the beholder.

PS. Bojan Bazelli ASC, (August'13 American Cinematographer) says that the new Kodak 5203 stock he used at 25 ISO on "The Lone Ranger" delivers 16-18 usable stops of latitude.

aka "expose for the shadows" rather than for highlights, right?
This is good to know but still, I guess in a high contrast situation, digging information in those shadows will be crucial and from that quick test with JB's files, I'm not that certain that it is really possible.

Have you tried downloading my original Prores files from the Pocketcam? The daytime stuff was at 400ASA so maybe try lifting the shadows on those and see if they get too noisy? (i haven't tried myself) Link to files is in the description :

Thanks a lot for sharing! I will definitely fiddle with those, too.

Thanks to all for your inputs in this discussion.
Sorry not to be able to be more technically specific, I'm still learning the DIY route

All I've done with those files is to use Premiere's 3-way color correction plug-in and play with the shadows wheel's input sliders (left and middle ones). All pictures posted are stills exported from Premiere 5.5.
If I'm doing anything wrong here, I'd be happy to learn how to do it properly!

Again, my intention is not to put this camera down at all, but instead I'd like to have a realistic understanding of what its limitations and strengths are.
One of the limitations of my hacked GH2 is that I usually have to under-expose to avoid clipping the highs, which makes it crucial to be able to recover details in the low and low-mid tones. As I tried to demonstrate with that GH2 rails shot, shadow details recovery is very limited. Below is one more example.

GH2 original, shot to protect the sky highlights:


Trying to recover shadow details in the wall and in the people's coats:


I am not willing to trade the articulated screen, vu-meters, in-camera edition/deletion, good battery life, EVF, stills capabitliy, etc. of the GH2 for a camera that would not help in the shadow details department, even though I understand its inherent DR qualities.

MX is 12bit, Dragon is 16bit

Dragon isn't finished or shipping yet.

And Epic is ????

This is a real names forum by the way.

jb

With my limited understanding of color correction, I've tried to fiddle with JB's ProRes Pocket files.


Overall, as soon as I touch the original, I find the noise to be distracting.

Most cameras stuggle with DIRECT sunshine on a building and DEEP shadows, which this scene has. it's aHUGE contrast range to try and capture.

The DR is large, but you still can't capture it all.....

Generally "video" looks better when the highlights aren't clipped....

You also didn't get any "banding" in the sky bending that grade did you ?

This is also a real names forum please.

jb

OK - thanks!
That totally makes sense and helps me understand better what can or can't be achieved with any digital camera, including this one.

@pask74
I grabbed a screenshot of your photo and got this in LR5.
Possibly the worst post scenario.
I am sure you can do better with your original files.

@pask74[/code]
I grabbed a screenshot of your photo and got this in LR5.
Possibly the worst post scenario.
I am sure you can do better with your original files.

Thanks for taking the time!
An experienced colorist would most probably do much better than me with this shot, although I wouldn't dare to go to your extremes (pink-ish murk etc.), to be honest.

The original Premiere export is attached, in case you'd be kind enough to play with it.
Thanks!

GH2 original Premiere export

cath dark.jpg (484.51 KiB) Viewed 15024 times

How about this, a bit less aggressive...(half a minute job).

…16 bits is still vastly more bit depth than 10 and 12

It's also worth noting that Alexa RAW is also 12-bit log internally derived from 16-bit linear. The methodology used by the BMCC is not exactly the same but is similar.