What exactly is dynamic range? Shooting RAW vs. regular video

[Chris Davis]

I have some basic questions about the difference is between shooting in RAW vs. traditional video, dynamic range in general, and implications for post.

Although I’m asking because I’m interested in the RED Scarlet, it may be easier to word my questions in DSLR terms (I’m not terribly knowledgeable about DSLRs, but I do know more about them than RED cameras).

1. When shooting RAW footage, is does the ISO setting I shoot with affect the way the actual red, green and blue photosites record the light, or is the ISO only metadata?

For example, lets say that a Canon 5D Mk. II shows about as much grain at ISO 1600 as does a Rebel EX does at ISO 400. If I were to shoot at ISO 200 with both a 5D Mk. II and a Rebel EX, would the 5D Mk. II give me an extra two stops of room to correct an underexposed shot?

2. How much more room do I have do I have to correct under or overexposed shots when using RAW rather than JPEG?

3. What exactly is dynamic range? When I think about it in video terms, I think of how over or unexposed a shot can be and still be correctable in post and keep most of the detail (e.g., how many stops overexposed can a shot before blowout). Or does it have more to do with how much really light and really dark information can be in a single shot, before some of the information is over or underexposed?

Thanks for your time,
Chris

 

[Graeme Nattress]

1. on a DSLR ISO effects the recorded RAW. On a RED, it does not as it's just metadata.
On a DSLR, because the gain is analogue, for every ISO stop you increase, you loose one stop of highlight dynamic range. On RED, the dynamic range is constant through ISO change. It does not decrease with increase of ISO.

2. On a DSLR, generally at least a stop or two extra highlight room is there in RAW over JPEG. On a RED it's ISO dependent. For each ISO stop you go up, you increase the headroom by a stop.

3. Latitude is by how many stops you can over or under expose a shot. But it's obviously dependent upon the range of the scene. Dynamic range is the range between light and dark that can be recorded in a single shot.

Graeme

 

[Chris Davis]

Thanks, Graeme. I'm a hobbyist, and I've been watching the prosumer cameras to see when to buy a prosumer camera. With the upcoming introduction of the RED Scarlet, it looks like time. I'm trying to understand the difference between working with Redcode RAW and regular video.

So my questions may seem quite basic. I've got some followup questions for you, or whoever else would like to help...

1. on a DSLR ISO effects the recorded RAW. On a RED, it does not as it's just metadata.
On a DSLR, because the gain is analogue, for every ISO stop you increase, you loose one stop of highlight dynamic range. On RED, the dynamic range is constant through ISO change. It does not decrease with increase of ISO.

What is "highlight dynamic range", and what does it mean for the gain of a DSLR to be "analogue." Does this mean that, on RED, I have the same ability to correct over or underexposed shots when shooting, say, ISO 100 as ISO 400, or would that be referring to "latitude." I'm not fulling this difference between DSLRs and RED. Could you please give some more detail here?

2. On a DSLR, generally at least a stop or two extra highlight room is there in RAW over JPEG. On a RED it's ISO dependent. For each ISO stop you go up, you increase the headroom by a stop.

What is "highlight room"? Does "headroom" refer to how much I can correct an over or underexposed shot in post (or, again, is that "latitude.")

Latitude is by how many stops you can over or under expose a shot. But it's obviously dependent upon the range of the scene. Dynamic range is the range between light and dark that can be recorded in a single shot.

I think understand you here. What determines how much dynamic range a camera has? Does shooting RAW help with dynamic range, or just with latitude? Does photosite size/sensitivity play a role?

Again, thanks for bearing with me on this,
Chris

 

[Graeme Nattress]

Dynamic range is a tricky subject and you ask good questions! By adjusting the brightness of the curve that takes the linear light data off the sensor and adjusts it to make it nicely viewable, we can adjust where the the middle of the dynamic range lies. In RED Cameras, this is done digitally to the data, after recording, not before, and is such that for each stop you go up in ISO, your mid grey value gets a stop darker, thus giving you one more stop extra dynamic range in the highlights (the distance between mid-grey and full clip) and one less in the shadows. You are just shifting the mid point up and down the scale.

In DSLRs this gain is applied using analogue circuitry before the signal is digitized off the sensor. This means that the gain is "burned in" before recording, and you actually loose a stop of highlight DR for each stop you go up.

To determine DR, shoot a chart. A 41 step (each step is 1/3 stop) Stouffer chart is pretty standard for this, but DSC make a nice 16 stop chart. You backlight the chart, expose the camera, and look for how many wedges you can see, and determine where you think the noise is more than the detail in the shadows to say where it stops at that end, and at the top end, you look for clipping.

Graeme

 

[Chris Davis]

...In RED Cameras, this is done digitally to the data, after recording, not before, and is such that for each stop you go up in ISO, your mid grey value gets a stop darker, thus giving you one more stop extra dynamic range in the highlights (the distance between mid-grey and full clip) and one less in the shadows. You are just shifting the mid point up and down the scale.

In DSLRs this gain is applied using analogue circuitry before the signal is digitized off the sensor. This means that the gain is "burned in" before recording, and you actually loose a stop of highlight DR for each stop you go up...

Got it, thanks.

Maybe I could ask about the RED ONE for reference since the Scarlet 2/3" is not out yet.

It seems that with regular video, you can only over/underexpose by one to two stops before the shot's unsalvageable in post.

But if I understand you correctly, shooting Redcode RAW totally changes this...

To clarify, say, on the RED ONE, say I shoot ISO 100, and underexpose a shot by fully four stops. Can I simply correct this in post by dialing in ISO 800 in post (thereby just adding a little grain but not losing much detail)? What about if I way overexpose by four stops while shooting ISO 800? Can I simply dial in ISO 100 in post and correct this, and maybe even take out some grain?

 

[Graeme Nattress]

Regular video is pre-processed and quite a bit of range is lost in that pre-processing. Shooting RAW video is like shooting RAW stills, in that such processing is for viewing only, and you can decide later on how to process for the best.

So, yes, you can do that, but I'd not recommend under-exposing by 4 stops. It's always best to get the best exposure you can

Graeme

 

[Chris Davis]

Regular video is pre-processed and quite a bit of range is lost in that pre-processing. Shooting RAW video is like shooting RAW stills, in that such processing is for viewing only, and you can decide later on how to process for the best.

So, yes, you can do that, but I'd not recommend under-exposing by 4 stops. It's always best to get the best exposure you can

Graeme

I used that example to try to better understand how ISO works with RED... Also, I should have said three stops, as that's the difference between ISO 100 and ISO 800.

1. If I hypothetically did overexpose by three stops while shooting ISO 800 with a RED ONE, then just correct to ISO 100 in post, would there be any difference visually between that and setting ISO 100 during the shoot and exposing correctly? Seems it should be the same thing if the aperture/shutter speed was the same, and ISO is just metadata.

2. Also, how much negative ISO/gain can the RED one deal with?

Again, thanks for your help here, just trying to better understand what I might expect from the Scarlet 2/3"...

 

[M Most]

1. If I hypothetically did overexpose by three stops while shooting ISO 800 with a RED ONE, then just correct to ISO 100 in post, would there be any difference visually between that and setting ISO 100 during the shoot and exposing correctly? Seems it should be the same thing if the aperture/shutter speed was the same, and ISO is just metadata.

2. Also, how much negative ISO/gain can the RED one deal with?

I think you're misunderstanding what Graeme is saying here. A simpler way of putting it is that the Red's sensor has fixed characteristics. There is no mechanism provided for changing those characteristics such as, for instance, changing the analog bias to boost the response (an ability of a number of DSLR's). There is an absolute minimum value to what the Red sensor can capture, and an absolute maximum value. All values in between are captured, and the range of these values is wide enough that, assuming sensible exposure, the lowest value will usually be lower than the lowest value in your scene, and with proper care taken, the highest value will be higher than the highest value in the scene. The ISO control lets you control the midrange of these values after capture, so it becomes possible to, say, overexpose a bit (i.e., "expose to the right" in Red parlance) and, in post, bring those values back to the middle of the range, giving you proper contrast, a pleasing image, and less noise (because as you go to higher exposure values, the noise component, which is constant, represents a much lower percentage of the overall value). Graeme is not saying that you can change the absolute minimum and absolute maximum values that the sensor can properly represent, because you can't. What you can do is balance the tradeoffs of excessive noise (a problem as the absolute values get lower) and white clipping (which results when you hit or exceed the absolute maximum value) by playing with ISO settings and moving the captured values up and down the histogram as desired.

Graeme, am I at least somewhat correct in my interpretation here?

 

[Graeme Nattress]

Yes, you're pretty good there. The sensor is what it is. What you're doing with ISO on RED is choosing the balance you need between highlight and shadow range.

We don't have any negative (or positive) gain at all in the sensor. We rate the sensor at ISO320 which leads to a pleasing, low noise image with reasonable highlight protection. Going lower ISO than that may mean to clip highlights early, leading to perhaps an undesirable look to the footage.

Graeme

 

[Chris Davis]

...We don't have any negative (or positive) gain at all in the sensor. We rate the sensor at ISO320 which leads to a pleasing, low noise image with reasonable highlight protection. Going lower ISO than that may mean to clip highlights early, leading to perhaps an undesirable look to the footage...

1. Perhaps I don't fully understand what gain is. I thought it was just another way of talking about ISO; I thought that 6 db of gain is the difference between ISO 100 and ISO 200, which, with RED is just metadata. Could you please explain what am I missing?

2. OK, so let me make sure I've got this right, if I may use another example... Say I were to shoot to ISO 400 with a RED ONE. If I underexposed the shot by 2 stops, I could correct in post to ISO 1600, although this would add considerable grain. If I overexposed by by two stops, I could dial in ISO 100 in post, but this would clip the highlights early. And if I overexposed by three stops, there would be no way to dial in negative ISO, so I'd be SOL. So I've got at most two stops of leeway to under/overexpose.

When I look at clips of the Canon XL H1 and the JVC HD 100 each at two stops over exposed, they look salvageable.
http://www.adamwilt.com/HD/4cams-part2-framegrabs/
(click "name," then "TX-14-HD100.jpeg" and "TX-14-HLH1.jpeg")

I had thought that, because of Redcode RAW, with a shot using the RED ONE, you could correct for much more overexposure in post than just two stops (which these regular video cameras seem to be able to handle). In other words it has much more latitude than these cameras that just shoot video. Is this accurate? Could you please give some detail here?

Again, just a hobbyist here trying to understand this stuff... Thanks for working with me here.

-Chris

 

[Graeme Nattress]

Definitely detail totally lost in those highlights, in the wall paper etc. Also, notice the color shift in the wall paper.

Graeme

 

[Chris Davis]

Definitely detail totally lost in those highlights, in the wall paper etc. Also, notice the color shift in the wall paper.

Graeme

So the RED ONE would look much better at ISO 100?

 

[David Mullen ASC]

There's correctable and then there's acceptable... you can correct two stops of overexposure but it all depends on the reflectance of the overexposed object - a black bowling ball that was two stop overexposed, no problem -- a white wedding dress, big problem. There is detail in a caucasian face overexposed by two stops on a digital camera, but there starts to be some oddities after you correct it to normal because the brightest highlights in the skin have become clipped.

Gain is an amplification of a signal. "Analog gain" is basically amplifying the signal from a sensor before it gets converted into a digital signal. The RED does not do any in-camera gain so basically you are playing around with the signal levels in post. The reason you might set your RED camera to 200 ASA or 400 ASA instead of 320 ASA, for example, is so that your monitor signal is corrected to appear normal at that ASA rating, and it tracks as metadata so that the files open up automatically with that correction. But you could chose to ignore the metadata, open up the files at 320 ASA, and then correct them manually.

All "RAW" means is that you are recording all the information the sensor captures, that's all. No processing tricks other than compression in the case of REDCODE. No conversion of the monochrome Bayer-filtered image into a viewable color (RGB) image, though the camera does make a real-time 720P conversion for monitor output. But that's not recorded.

In theory, you could design a RAW camera that had poor dynamic range. The term is not necessarily a guarantee of anything other than a lack of processing.

Some video cameras use knee compression to squeeze more highlight information into the narrower gamma for viewing on a broadcast monitor. You could do the same to a RAW recording in post if you wanted to.

The dynamic range of the RED camera's RAW recording is good, but I wouldn't say that it's miles ahead of other digital cameras, or radically different than other digital cameras, and it's certainly nowhere near what color negative film captures. You may find that it has a stop or two more range than a typical video camera though.

The point of RAW recording is two-fold -- one is that because the monochrome Bayer-filtered image has not been converted to RGB, it's naturally a smaller-sized file to record compared to the same pixel resolution tripled in data simply by being converted to RGB (4K RAW is one-third the data of 4K RGB). The second is to leave digital processing decisions to you in post.

---

The RED ONE is basically around 320 ASA if you want to use film terms like ASA. This rating basically gives you a good balance of highlight and shadow detail. If you want to use a slower rating and essentially "overexpose" the sensor, you would be biasing the recording towards the shadows, capturing more information there but losing more highlight information. And you'd also be reducing the noise level.

So it's always a battle between noise vs. avoiding clipping highlights, which is digital's Achilles Heel. It cuts off overexposed detail suddenly and dramatically at the clip point. Lower ASA ratings, less noise, worse clipping -- higher ASA ratings, worse noise, less clipping. Which is why 320 ASA is more or less a sweet spot between the two ends.

However, if your image is low in contrast without many bright highlights, you have more flexibility to use the lower-than-320 rating in order to reduce noise. If the image is high in contrast with a lot of bright highlights, you have the option of using a higher-than-320 rating to reduce clipping at the expense of getting more noise.

 

[Chris Davis]

I think you're misunderstanding what Graeme is saying here. A simpler way of putting it is that the Red's sensor has fixed characteristics. There is no mechanism provided for changing those characteristics such as, for instance, changing the analog bias to boost the response (an ability of a number of DSLR's). There is an absolute minimum value to what the Red sensor can capture, and an absolute maximum value. All values in between are captured, and the range of these values is wide enough that, assuming sensible exposure, the lowest value will usually be lower than the lowest value in your scene, and with proper care taken, the highest value will be higher than the highest value in the scene. The ISO control lets you control the midrange of these values after capture, so it becomes possible to, say, overexpose a bit (i.e., "expose to the right" in Red parlance) and, in post, bring those values back to the middle of the range, giving you proper contrast, a pleasing image, and less noise (because as you go to higher exposure values, the noise component, which is constant, represents a much lower percentage of the overall value). Graeme is not saying that you can change the absolute minimum and absolute maximum values that the sensor can properly represent, because you can't. What you can do is balance the tradeoffs of excessive noise (a problem as the absolute values get lower) and white clipping (which results when you hit or exceed the absolute maximum value) by playing with ISO settings and moving the captured values up and down the histogram as desired.

Graeme, am I at least somewhat correct in my interpretation here?

Thanks, mmost. I missed your post because I didn't look back. Can negative ISO be dialed in on post to correct for an underexposed shot (albeit with clipping of highlights, etc.), in a way that would give much better results than working with regular video?

Most of the people here are pros who can ride the aperture, focus and zoom at the same time in the middle of the shot. I'm lucky just to get optimal exposure on a static shot. So what I'm really wondering is how many more stops of room do I have to screw up my shot when working with RED over video?:thumbsup:

David: Thanks for your post. Running out of time, but will read your post more carefully and get back.
Thanks,
Chris

 

[David Mullen ASC]

Most of the people here are pros who can ride the aperture, focus and zoom at the same time in the middle of the shot. I'm lucky just to get optimal exposure on a static shot. So what I'm really wondering is how many more stops of room do I have to screw up my shot when working with RED over video?:thumbsup:

Depends on the video camera. If you are talking about a prosumer video camera, maybe a stop or two more information with the RED to play with in post, depending on whether your camera was using knee compression.

The thing is that some prosumer cameras allow you to play around with the gamma to squeeze more information into the range that gets recorded, but there's no free lunch, often it creates other problems like noise in the shadows or oddly greyed or tinted whites, etc.

"Latitude" means many things, it's not just about how many stops of exposure you have recorded, it's about your ability to make corrections. A prosumer camera often is using some sort of heavy compression scheme and color subsampling that will limit your ability to correct it in post without getting artifacts. So even if the RED camera's 12-bit RAW recording only had one more stop of exposure information in it (and it might have more), you'll probably find it easier to manipulate in post after conversion to RGB than the 8-bit 4:2:0 recording of an HDV camera, for example.

 

[Graeme Nattress]

As David says, that's an un-answerable question. What we do is ensure that you get as much of what the sensor sees as you can recorded, and we don't do processing decisions in camera that reduce that.

Graeme

 

[David Mullen ASC]

A professional camera like the RED ONE is not designed to be more idiot-proof, so to speak, than a consumer camera -- it's intended to be shot as "correctly" as possible or as desired. If you have trouble exposing a scene properly, you're always going to have issues whether you shoot digital or with film, whether you use a consumer or a professional camera. So don't get a RED camera because you think you can get away with sloppier exposure techniques. The camera uses manual lenses, no auto exposure, just some exposure guidelines in the viewfinder image to help you.

 

[Chris Davis]

David. Thanks for the detailed responses...

...The dynamic range of the RED camera's RAW recording is good, but I wouldn't say that it's miles ahead of other digital cameras, or radically different than other digital cameras, and it's certainly nowhere near what color negative film captures. You may find that it has a stop or two more range than a typical video camera though.

Depends on the video camera. If you are talking about a prosumer video camera, maybe a stop or two more information with the RED to play with in post, depending on whether your camera was using knee compression.. A prosumer camera often is using some sort of heavy compression scheme and color subsampling that will limit your ability to correct it in post without getting artifacts. So even if the RED camera's 12-bit RAW recording only had one more stop of exposure information in it (and it might have more), you'll probably find it easier to manipulate in post after conversion to RGB than the 8-bit 4:2:0 recording of an HDV camera, for example.

When you say, “a stop or two more information,” do you mean a stop or two to both under and over-expose (total of 2-4 stops more latitude)?
Also, as far as the Scarlet 2/3” is concerned, can we assume that it would have a similar latitude as the RED ONE, or would the fact that the individual photosites are about 1/3 the area of the those on RED ONE possibly reduce the lattidude?

So don't get a RED camera because you think you can get away with sloppier exposure techniques. The camera uses manual lenses, no auto exposure, just some exposure guidelines in the viewfinder image to help you.

I sincerely appreciate the advice – I had planned to look at other prosumer options as well as the RED Scarlet 2/3”. Hopefully the other manufactures will start offering us real codecs (like AVC-Intra) in prosumer cameras in response to the RED Scarlet.

I was inquiring about the RED ONE rather than the Scarlet for reference because it's the only RED camera that's been released, but I would be interested only in the Scarlet 2/3" fixed lens. I had read on this forum that Jannard said that the Scarlet 2/3” was going to have manual everything, though I didn’t see the link.

 

[Gavin Greenwalt]

The easiest way to think of metadata for RED and ISO is to think of it as levels in Photoshop.

You load the file. It is what it is. From that you can adjust your exposure by adjust levels and gamma--but just like on a histogram or levels in Photoshop you're clipping data on one end or the other.

 

[David Mullen ASC]

When you say, “a stop or two more information,” do you mean a stop or two to both under and over-expose (total of 2-4 stops more latitude)?
Also, as far as the Scarlet 2/3” is concerned, can we assume that it would have a similar latitude as the RED ONE, or would the fact that the individual photosites are about 1/3 the area of the those on RED ONE possibly reduce the lattidude?

I mean what I said, a stop or two more information -- whether you want to use that to record more shadow information or more overexposure information is up to you.

The Scarlet isn't out yet so we have no idea, but I suspect that the 2/3" version will be similar to the current RED ONE in terms of dynamic range, the Mysterium-X version even more, and the Monstro version even more than the Mysterium-X, being a 14-bit processor I believe. But who knows.

Believe me, one more stop of information is significant, nothing to sneeze at.