Brawleys first DNGs from the BMPCC

[Axel]

Here, used lens: SLR magic 12mm. Funny colors in parc sky are CA. Or what?

[Ben Prater]

I played with images for a few minutes in Adobe RAW. Lots of color information to play with. Looks great. However, I see a little color fringing in certain parts of the images. For instance, in the trees where the sky peaks through. Are these lens artifacts? Or does the sensor have really gnarly highlight rolloff?

 

[jgharding]

From what I can gather so far, the highlight rolloff is more like a cliff jump, and is gnarlier than Waynes World

[Nikkor]

There are like 3.2 stops in the highlights you can recover, but the midtones are very noisy. I guess they somehow moved everything to the right (does that make any sense to the sage around here?)

 

It's curious that the previews on the download page look like they had the actual exposure, but then in photoshop it displays 3 stops higher.

 

Here two examples:

 

the first one appears totally exposed to the right when you open it in Photoshop, there are two blown spots, one on the spherical street light and one on the left side under the wild life sign.

 

 

second: appears like a normal metered photo in PS but in fact has noise like a recovered underexposed one

 

[jgharding]

I suppose they kind of "baked in" the ETTR rule.

 

Alexa beats most of the rest due a complex reading system that works something like Magic Lantern 5D MKiii dual ISO...

 

DGA simultaneously provides two separate read-out paths from each pixel with different amplification. The first path contains the regular, highly amplified signal. The second path contains a signal with lower amplification, to capture the information that is clipped in the first path. Both paths feed into the cameras A/D converters, delivering a 14 bit image for each path. These ima-ges are then combined into a single 16 bit high dynamic range image. This method enhances low light performance and pre-vents the highlights from being clipped, thereby significantly extending the dynamic range of the image.

ALEXAs sensor design provides 32 pairs of outputs. Each channel is divided into a high amplification (gain) path (H) and a low gain path (L), resulting in 64 channels arriving at the 14 bit A/D converters.

In the final images, the shadow areas are re-constructed from the high gain path and the highlights are reconstructed from the low gain path for an image containing meaningful luminance information in all 16 bits.

[Andrew Reid]

Don't think the mids are noisy, I think the shadows when pushed into the mids are noisy. John has exposed for the sky I think. These shots are beautiful looking, and really push the boundaries of 13 stops dynamic range. This is a 10x better image than what we're used to from the GH2 and GH3.

[Nikkor]

It certainly looks worse than ML raw from the 50D.

[Andrew Reid]

It certainly looks worse than ML raw from the 50D.

 

http://www.specsavers.co.uk

[Nikkor]

Yeah sure Andrew, bla bla bla

 

No adjustment, maybe midtowns are shadows for you... the shadows btw are horrible...

[jgharding]

Maybe the mids are noisy because they were once shadows, and it's pushing out of camera, hence the small overexposure latitude?

 

Unlike the alexa, which uses a low-gain pass for the shadows, this and most other cameras push the single sensor pass to boost dark areas. (I don't know about the high-end Sonys yet).

 

Effectively this means applying ISO (gain) to the underexposed areas of the sensor, so naturally they become noisy.

 

Arris really nailed it with Alexa, maintaining something like 7 stops over, and the clean shadows, using those two gain settings at once, reading the sensor with double channels.

 

This is what ML dual ISO on 5D MKiii mimics with some issues.

 

Maybe the 70D double photodiode sensor could be hacked to do what Alexa does? Let's hope so... It would depend on the number of channels off the sensor, so probably not.

 

Those who've done a lot of raw photography will notice this noise from when they've boosted shadows to the middle, suddenly those areas have a lot of gain, so a lot of noise.

 

In short what I'm suggesting is the channels that read the sensor are already reading a single that's gained quite a bit, so what read as mids, were once the shadows, and the sensor isn't quite as sensitive as it should be?

 

That might explain the low overexposure latitude. It's an issue with Red Epic to, that's much noisier than Alexa.

 

Just a hypothesis though, I've got no scientific method to back it up, just bits of knowledge I'm trying to tie together.

 

What's truly wonderful is that the Amira may come in at 20k, and that's basically an Alexa.

 

Then the C300 can f*** right off.

 

But yes for the money this is great. I do love and prefer the 5D MKiii raw now I've started playing with it though, and prefer it, but the colours really need taming in that. Horses and courses etc!

[Nikkor]

Nah, it's not only the common noise, you can see there are a lot of debayer errors also the sky has a continuous pattern, it's noise plus combining pixels in a fast way. Actually when you make a camera that downsamples you should use that to reduce noise, not to get more.

[jgharding]

What do debayer errors look like? What do you mean combining pixels? I'm keen to learn a little more.

[Andrew Reid]

Everybody can break an image, much harder to make one.

 

This amount of luma noise is pretty normal and you can get rid of it without smearing fine detail. Try that instead of just breaking stuff. See if you can fix it. If I have time I will fix it and show you.

 

These are silhouetted areas of the image and under exposed. I have opened the DNG in Photoshop and that's what they are.

 

Also with raw you need to apply any noise reduction in post as the camera doesn't do any. You have not applied any NR and you've blown the area up 400% or something to magnify the grain. The grain is a nice fine texture on the BMPCC and nothing like what you're representing.

 

Fine grain is actually quite a lot like film, and people often add it voluntarily in Film Convert so a bit of texture to the image isn't actually always a bad thing.

[Nikkor]

You are so smart Andrew, a daylight shot with a good sensor doesn't need any noise reduction, maybe these samples are just at some ISO they are not meant to be. The ship is definitely not a silhouette area, it's not shot against the sun altough it is in the "shadow", it's shot 90º off the sun with a lot of ambient light, looking east (if that is australia) at mid day, there should be no fucking noise at all. 

But hey, for vimeo videos, who cares... Let's wait for some lowlight samples, and the ETTR mystery solved.

[chrisso]

Don't forget this is beta testing an unreleased raw format. Don't write it off yet.

Yes, it is downtown Sydney. The sky will be unbelievable bright and there will be a huge latitude between sky and dark shadows.

[Axel]

The ship is definitely not a silhouette area, it's not shot against the sun altough it is in the "shadow", it's shot 90º off the sun with a lot of ambient light, looking east (if that is australia) at mid day, there should be no fucking noise at all. 

But hey, for vimeo videos, who cares... Let's wait for some lowlight samples, and the ETTR mystery solved.

 

I say, let's wait for some moving images. Noise is hard to evaluate from a single image. Yes, if it was a stills camera ...

 

It may be bad, it may actually look good. You know, all the high bitrate - small GOP hacks of the GH2 never added any real detail to the images, they dithered obscure parts in a way like film did, creating so called 'temporal samples'.

 

Yes, I know, iso noise is fundamentally different from film grain. But, like Andrew wrote, it still is something people add voluntarily to make their videos look more organic (they add random film grain simulation, which like iso noise, has nothing to do with the pixels in the image, it's an artificial layer).

 

If it really turns out to be an 'issue', there is a good chance that tools like Neat (or the pro version of Resolve?) can fix it easily.

 

Raw photography is also not free from noise. Adobe Raw within Photoshop has it's own NR filter. The noise you detected on the ship is nothing I would have found in original size (1080 viewed on a 1080 monitor from a viewing distance appropriate to, er, cinema), my guess is that it is negligible for video in most situations. In this photo, the noise disappeares completely with a mild 18% luma NR (visible in 400%), no big deal, since one has to treat the raw with several agents anyway.

 

What is more, the sensor is said to be comparable to that of the BMCC. The first Brawley demos also showed some noise, but it isn't discussed as problem anymore, afaik. Maybe it can be avoided altogether, maybe not. No reason to despair.

 

Paramount for me, I like the images, I am not for missing the wood for the trees. Or worse, to ignore the path underneath your feet! 

[gloopglop]

Hello friends, I am new here, long time lurker. Thank you sincerely for the discourse; it's extremely educational to me.

 

After a brief analysis of these .dng files in ACR my reaction is, in a word...pleasure~! These images feel naturalistic to me in a way that the BMPCC prores for the most part does not, not to shit on the prores and say it's worthless––it's very fine. But there is a certain character to these that I far prefer, perhaps because of the added effective dynamic range [when properly processed]? I'm really not sure; I appreciate all of your comments as they help me get a grasp on what is going on here. I would like to understand the 'why' of digital cinema as much as possible and ultimately I am just a beginner.

 

Anyway, again, ACR produced pleasing results for me with minimal effort, and a little noise reduction goes a long way if that's the look you're going for. I didn't find what I would consider to be an unexpected amount of noise in any part of the images based on my perception of the exposure and lighting conditions.

 

I'm looking forward to seeing more raw imagery from the BMPCC and I look forward to seeing the analysis here on this board, thank you guys.

 

Cheers!

[AKH]

One thing to note.  ACR applies sharpening by default on dng files.  So if you haven't zeroed it the noise does get lifted a bit.

[Axel]

One thing to note.  ACR applies sharpening by default on dng files.  So if you haven't zeroed it the noise does get lifted a bit.

 

You are right. It is a default Photoshop preset applied to every raw file. And setting it from "25" to "0" practically eliminates what is described as "noise". Keep in mind, that camcorders with intelligent compression algorithms also develop raw images under the hood, that they add gain, NR and sharpening. But sharpening must be applied to the BMPCC images, obviously. Perhaps there are better methods (.i.e. in Resolve, as a last step of grading then, there are also discussions everywhere whether to apply NR first or last, plainly the right answer is first, the node model makes it easy to get the order right - only that there is no denoiser in Lite :( ), but I guess it just needs some understanding on how the tools work:

 

When opening a camera raw image file, the Camera Raw plug-in calculates the threshold to use based on camera model, ISO, and exposure compensation.

 

 

This is from here. Hystery begone!

[hmcindie]

You can actually even use ACR on 8-bit jpeg files. You'd be surprised by how well it can bring back highlights (it does an intelligent recovery using all channels). It also sharpens nicely.

 

A lot of people impressed by the "sharpness" of the 5dmarkIII raw files are actually impressed by ACR.