I agree. Man....if we can master the workflow, you're looking at somewhere around 14 stops (as far as I can tell, please chime in if you know otherwise). Wowser....
Canon Hack (Magic Lantern) brings HDR video to T2i and others
- Thread starter HHL
- Start date
I agree. Man....if we can master the workflow, you're looking at somewhere around 14 stops (as far as I can tell, please chime in if you know otherwise). Wowser....
Jarrett P. Morgan
Well-known member
Yeah! I heard that eariler versions couldn't handle motion really well, but from the looks of that last video, the interpolation seemed to work pretty well with the water, cars, and focus pulls (all things I would have thought would trip up the interpolation)
This is definitly something to mess with and try and master. wow.....
Kevin Marshall
Active member
I may get ML for my T2i to check this out, but I certainly have my reservations...
Motion interpolation. From what I understand form the explanation in the video, you have essentially two streams of 12fps video, and the other frames for each stream are interpolated. I guess they can use the existing (though differently exposed) alternate frames to aid in the interpolation, but I'm still a little wary here at the outset.
ISO. Their method of attaining "different" exposures is changing ISO. This does avoid artifacts from changing aperture or shutter speed, but has it's own (potentially crippling) issues. The first is seen in most of the test videos - noise. By using a higher ISO for your base image (to allow for a lower one for the underexposed frames), you're raising the overall noise level of the image. You could use noise-reduction software (further complicating the workflow and increasing work time), or you could limit yourself to lesser ISOs (limiting the usefulness of the feature), but both of these are potentially heavy compromises.
Beyond the noise - or rather at the root of it - there is the fundamental issue that changing ISO isn't changing the amount of light either hitting the sensor or being read by it. It is changing the amount of gain applied to the image after the fact. A saturated photosite isn't going to become unsaturated because of a lower ISO. So to retain highlights, you're still essentially underexposing your image enough to avoid clipping, and then raising the gain in another stream to bring the shadows back up.
The point is this: you're still limited by the native DR of the sensor. The hack just applies different gains and gamma curves to the same data coming off the sensor to try and maximize that range. So I think the theorization of 14 stops is definitely premature, if not ridiculous. with this method of HDR, it is not possible to get a any more DR out of the camera than a very well-processed RAW frame from the stills mode.
What I think works better in this situation is going for a flatter gamma curve in the vein of Cinestyle. It has the potential for the same DR as the ISO method, but without all of the interpolation and blending in post.
That said, there is the possibility of some advantage to splitting that data across two 8-bit files rather than one, but it remains to be seen (and likely is a subjective decision) whether that gain exists beyond theory, and if it (in practice) outweighs the other compromises of noise and frame interpolation. I'm interested in checking this out and testing these concerns - I may just bite the bullet and put Magic Lantern on my T2i.
Motion interpolation. From what I understand form the explanation in the video, you have essentially two streams of 12fps video, and the other frames for each stream are interpolated. I guess they can use the existing (though differently exposed) alternate frames to aid in the interpolation, but I'm still a little wary here at the outset.
ISO. Their method of attaining "different" exposures is changing ISO. This does avoid artifacts from changing aperture or shutter speed, but has it's own (potentially crippling) issues. The first is seen in most of the test videos - noise. By using a higher ISO for your base image (to allow for a lower one for the underexposed frames), you're raising the overall noise level of the image. You could use noise-reduction software (further complicating the workflow and increasing work time), or you could limit yourself to lesser ISOs (limiting the usefulness of the feature), but both of these are potentially heavy compromises.
Beyond the noise - or rather at the root of it - there is the fundamental issue that changing ISO isn't changing the amount of light either hitting the sensor or being read by it. It is changing the amount of gain applied to the image after the fact. A saturated photosite isn't going to become unsaturated because of a lower ISO. So to retain highlights, you're still essentially underexposing your image enough to avoid clipping, and then raising the gain in another stream to bring the shadows back up.
The point is this: you're still limited by the native DR of the sensor. The hack just applies different gains and gamma curves to the same data coming off the sensor to try and maximize that range. So I think the theorization of 14 stops is definitely premature, if not ridiculous. with this method of HDR, it is not possible to get a any more DR out of the camera than a very well-processed RAW frame from the stills mode.
What I think works better in this situation is going for a flatter gamma curve in the vein of Cinestyle. It has the potential for the same DR as the ISO method, but without all of the interpolation and blending in post.
That said, there is the possibility of some advantage to splitting that data across two 8-bit files rather than one, but it remains to be seen (and likely is a subjective decision) whether that gain exists beyond theory, and if it (in practice) outweighs the other compromises of noise and frame interpolation. I'm interested in checking this out and testing these concerns - I may just bite the bullet and put Magic Lantern on my T2i.
TrueIndigo
Well-known member
I don't use my 550D much since getting a 60D (I really miss the articulated screen) so I think I'll finally start experimenting with ML on the 550D. And the thing that has finally jolted me into action has been this HDR feature! The other ML features are very interesting and useful, but this latest option has really captured my imagination. It's still too early to tell, but considering the cost of the equipment, could you ask more from 8-bit video that has highlight protection like this! I'm also intrigued by what this hack means for the next round of Canon DSLRs next year...
By the way folks....
Something to keep in mind, by using a beam splitter, any cam could do this. The workflow has been the 'missing link' for me at least. I just might have to pull out the beam splitter and give the workflow a try some time in the upcoming weeks as the workflow improves.
Something to keep in mind, by using a beam splitter, any cam could do this. The workflow has been the 'missing link' for me at least. I just might have to pull out the beam splitter and give the workflow a try some time in the upcoming weeks as the workflow improves.
Doesn't Red's HDRx essentially use two streams?
Jarrett P. Morgan
Well-known member
The RED HDRx uses two frames per frame, one with a normal shutter speed and one with a shorted one, thus protecting for the highlights.
I do not know what the post workflow is on HDRx though, just the concept behind it.
Isn't that hat the Magic Lantern hack essentially does, thus resulting in an overall halving of frames rates?
Kevin Marshall
Active member
Read my earlier post.
ISO isn't like it is in film - where there are stocks with different sensitivity. You can't change the sensor's sensitivity, you can just change the amount of light hitting it (aperture, ND), how long it's hitting it (shutter speed), or the amount of gain applied afterward (ISO, gain). For any given exposure, there is a fixed amount of light hitting the sensor (and being read off by it). If you're clipping at the sensor level (not because of the gamma curve applied), you can't gain that down into usable information. You can gain the shadows up further, though that comes at the cost of noise.
Either way, ISO changes don't change your DR. It's still using the same base exposure, just with different gains applied. You could get the same effect with a flatter profile (like Log-C, S-Log, etc), and keep full frame rate and a single stream. It also could be done with RAW.
Making different "exposures" without actually changing how much light is hitting the sensor (or being read off of it) isn't getting any more DR than the sensor is capable of, it's just a roundabout way of maximizing a single exposure.
Sort of. Red's HDRx does utilize two streams of video, which limit your available frame rates approximately by half. However, there are a few important differences.
One - HDRx (on the Epic, and on 3K and below on the Scarlet), still allows you to have full 24fps per stream. I guess if you could do 48fps in 720p with ML, it would help alleviate this - but it brings me to my next point:
Time difference. Even if you did 48fps with the ML HDR, each frame would still be offset, and not from the same point in time, leading to artifacts. HDRx works differently, though yielding it's own artifacts. HDRx captures both frames at the same time, though one is a much shorter exposure than the other (depending on the number of stops difference). This juxtaposes the short, sharp exposure of the x-frame with the motion blur of the a-frame. The effect is amplified the more difference there is in the exposures.
There is also the difference between using ISO vs shutter, which is already detailed throughout my post.
Last edited:
Hey Kevin. Thanks for your input. I think that there is a misunderstanding here about what is going on and what is purported as a gain in DR.
No question, the DR of ***THE SENSOR*** is ***NOT*** improved with this method. I'm so sorry if anyone thought that it would help the DR of the sensor.
Also not a question (from posted images), the DR of ***THE FINAL VIDEO*** IS CERTAINLY improved.
It's a very simple theory that has ***nothing*** to do with the sensor. It is exactly the same principle as HDR photography:
Combine multiple exposures at different settings to reach further (more DR) than the sensor can by its self. (in a single exposure)
As for it being the same of Red. No, it is not (as you have pointed out). I hope this helps.
Here's another example:
ps...I can't tell you how much I wish we could get this sort of range from this sensor....I wouldn't even care about the HDR video principle if that were possible. Sadly, it cannot, as such, we're stuck getting multiple exposures to get this sort of range.
Notice how the building (on the right and the left for that matter) is completely clipped, yet on the HDR video it is completely present. Great example of added range to the final video. (not to the sensor as you have properly pointed out)
ps...I can't tell you how much I wish we could get this sort of range from this sensor....I wouldn't even care about the HDR video principle if that were possible. Sadly, it cannot, as such, we're stuck getting multiple exposures to get this sort of range.
Notice how the building (on the right and the left for that matter) is completely clipped, yet on the HDR video it is completely present. Great example of added range to the final video. (not to the sensor as you have properly pointed out)
Last edited:
Kevin Marshall
Active member
I certainly didn't mean to impugn any of the hard work and ingenuity that went into the development of this feature. I just wanted to respond to some questions and control some expectations from some enthusiastic people.
It is true that I have a difference in opinion about maximizing the DR, but my experience is minimal with DSLRs, and non-existent with Magic Lantern - so the experience with other camera systems and the application of theory may be muddied by certain realities of this camera system that I'm not as familiar with.
After Christmas, I would very much like to test this feature, and challenge my experience and knowledge. Hopefully I'll have something to report.
Not at all Kevin. You weren't being negative or disrespectful. You're completely fine. Thanks again for your input. Have fun testing after Christmas! Cheers!