Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

HDRx can't cope with flames

paulherrin said:
give me the r3d and i'll give you a shot you can use.
Click to expand...

Paul, thank you for the offer, but this was just a test shoot.

When we film it in a few weeks time I will take onboard all the great advice given here, and the lessons learnt.

Out of interest, how would you have approached it?
 
compositing...

attachment.php
 
I got excellent results in the Flame. I brought in both tracks in 16 bit linear but I used the ISO control of each to compensate for the exposure difference. Then I used a logic op 'Max' to combine them into single 16bit HDRI image. That gave me the highlights from the x-track only where the sensor clipped in the primary exposure. It was a simple matter of color correcting to taste afterwards.
Magic motion was not needed on most of the shots.
Hint: You want to set the ISO of the X-track a little lower than the nominal exposure difference so that you don't get added noise from the two exposures.
This should be the default method of combining in red cine x. The simple mix really defeats the purpose of having two exposures because it leaves the clipped image in the result. It's a little shocking to me how many programs 'support' HDRx without a clue about imaging science.
Sam Edwards
 
For what I see, when you are dealing with HDRx, you have to blend your overexposed image just enough to recover your dark areas but without exaggeration, so it looks natural. then apply contrast to the whole thing..
 
Karim D. Ghantous said:
This would explain why I read (somewhere) that film is generally preferred for explosions. I wonder how you would have coped if you were forced to use video cameras?
Click to expand...
Underexpose about 2/3 of a stop to protect the highlights, and use lots of fill for the blacks. And do tests prior to the shoot if possible.
 
Pablo Moreno said:
F!@#$, that is awesome brother!, I wish I could see that in motion..
Click to expand...

me too... haha

playing with fire is fun, maybe antony will send me the whole shot or the 'real' one, i've already got the comp setup and ready to rock... obviously there are some details to tweak in finishing, but it's basically plug and play.
 
Pablo Moreno said:
Can you explain how it works?, is HDRx 48fps alternate, or it works some other way??
Click to expand...

Antony M said:
Essentially it works by reading the sensor twice. For example, if you are shooting a 2-stop HDR at 25fps with a 1/50 second exposure, then a reading is also taken after 1/200 second.
Click to expand...

From technical perspective the act of "reading" a sensor is rather complex. As the light starts entering the photosites of the sensor, the individual photons are captured in the "pixel chambers" (for the lack of a better word) building up a photovoltaic charge.
To "read" the sensor these have to be discharged and the resulting voltage is measured - generating a wavelength where the points of the curve represent the consecutive pixel charge values. The ASIC knows that individual pixels have colored filters above them to only allow light of a certain color range in (the Bayer pattern) and then converts this charge into a color information generating an image for the preview. The actual curve is saved as the RAW data - this is why the RAW compression is very different from compression applied directly to an interpreted image...
The important part of this process to understand is the fact that the act or "reading" the sensor actually resets all the pixel values - which is why this cycle is called READ-RESET...

On your average camera that shoots HDR (such as bracketed DSLR exposures) - the individual exposures start and are read at a different times - this is what creates problems with fast moving subjects...
On RED the two exposures actually start at the same time - they just finish apart, one sooner to capture the information of the highlights before they are blown out by the longer exposure aimed at the lower 2/3 of the image...
You don't have to be a rocket scientist to figure out how this is possible and how similar results could be achieved in post with bracketed exposures.
The key to make this work well is the extremely fast READ-RESET cycle - which also reduced the rolling shutter effect.

In any case you will see that the highlights are actually at the leading edge of the longer exposure - not "outside" in a different place...
However - the faster the motion of the subject - the more area it covers in the longer exposure (motion blur) and the more dramatic the distance from the leading edge to the "center" of the blur.

If you wanted to have the highlights in the center of the blur - a three exposures would be necessary - reducing the frame rates to 1/3 and tripling the data rate per frame of footage...
So let's say you wanted a HDRx of the setup Antony has mentioned - 1/50s with +2 HDRx...

The "normal" HDRx would expose as follows:
1st exposure - 1/200s
READ-RESET
2nd exposure - continues for 3/200s and is combined with the 1st 1/200 exposure to create a 4/200 second exposure, which is 1/50 of a sec...
So the A-track has 1/50 exposure and the X-track has 1/200 exposure...

To get the highlights in the middle you would need three exposures as follows:
1st exposure - 3/400s
READ-RESET
2nd exposure - 1/200s
READ-RESET
3rd exposure - 3/400s

The "normal" A-track will be created by combining (adding) all three exposures - 3/400 + 1/200 (which is 2/400) + 3/400 = 8/400, or 1/50s
The "highlight" X-track will be created by using the data from only the 2nd exposure - 1/200s
But the resulting HDRx will have the highlights in the center of the blur (basically in the same place), not at the leading edge as it is now with the current HDRx...

To make this work the READ-RESET cycle has to be as fast as technically possible - in a ideal world instant - and you would need much faster media that will support at least 1.5x the speed of the current media...
RED's READ-RESET cycles are already amazingly fast - this is what allows the build 4's long exposures by combining up to 16 frames...

Hope this helps...

Peter
 
i actually haven't ever had a shot yet where i needed hdr-x, but i haven't been shooting a whole lot lately. there are certainly situations that call for it, but i can't imagine a normal instance where you would need more than the 2 exposures other than a plug and play holy grail timelapse... or if you're strapped to a spaceship sailing towards the sun :)

i think the real solution is harnessing optical flow motion estimation (twixtor/reelsmart/kronos) augmented by a guide frame (i.e., the x frame looks at the a frame and 'matches it) to get an hdr image ready for tonemapping - otherwise it's just age-old compositing tricks...

or you could just do something crazy like develop a sensor with 18+ stops of dynamic range... hehehe :)
 
paulherrin said:
...or you could just do something crazy like develop a sensor with 18+ stops of dynamic range... hehehe :)
Click to expand...

I would still want a HDRx with it!
I do those crazy timelapses all the time and even the current +6 HDRx can't cope with the difference in light from down to after sunrise and vice versa during sunset. Working currently on a motorized aperture control with super slow, yet smooth action to deal with this...
 
awesome, that's the way to go... jumping down and fixing it in post sucks - i was doing some of that yesterday for a sun up & down shot :willy_nilly:
 
paulherrin said:
me too... haha

playing with fire is fun, maybe antony will send me the whole shot or the 'real' one, i've already got the comp setup and ready to rock... obviously there are some details to tweak in finishing, but it's basically plug and play.
Click to expand...

I'm more than happy to.... but how do you want me to get a whole R3D to you?
Most of the file share sites I use have a file size limit...
 
paulherrin said:
i actually haven't ever had a shot yet where i needed hdr-x, but i haven't been shooting a whole lot lately. there are certainly situations that call for it, but i can't imagine a normal instance where you would need more than the 2 exposures other than a plug and play holy grail timelapse... or if you're strapped to a spaceship sailing towards the sun :)

i think the real solution is harnessing optical flow motion estimation (twixtor/reelsmart/kronos) augmented by a guide frame (i.e., the x frame looks at the a frame and 'matches it) to get an hdr image ready for tonemapping - otherwise it's just age-old compositing tricks...

or you could just do something crazy like develop a sensor with 18+ stops of dynamic range... hehehe :)
Click to expand...

Actually, yes, I agree, somebody has to come up with some kind of sensor idea, that does automatically what we do manually.. like below..
 

Attachments

  • HotelLight.jpg
    HotelLight.jpg
    55.4 KB · Views: 0
  • HotelDark.jpg
    HotelDark.jpg
    39.1 KB · Views: 0
  • HotelHDR.jpg
    HotelHDR.jpg
    36.2 KB · Views: 0
Peter Majtan said:
I would still want a HDRx with it!
I do those crazy timelapses all the time and even the current +6 HDRx can't cope with the difference in light from down to after sunrise and vice versa during sunset. Working currently on a motorized aperture control with super slow, yet smooth action to deal with this...
Click to expand...

You can try this too, is very interesting..

http://19lights.com/wp/tutorials/hdr-timelapse/
 
Peter Majtan said:
Sorry Pablo - but no SW solution/fix can magically recover blown out image...
You either increase the DR of the sensor - or you have to control the aperture physically...
Click to expand...

I know that, but as far as I understand, the plugin works with two images, (over and underexposed), did you watch the other videos?
 
Last edited:
paulherrin said:
i think the real solution is harnessing optical flow motion estimation (twixtor/reelsmart/kronos) augmented by a guide frame (i.e., the x frame looks at the a frame and 'matches it) to get an hdr image ready for tonemapping - otherwise it's just age-old compositing tricks...
Click to expand...
No, the real solution is getting somebody who knows how to light the scene and expose it for maximum results. Everybody here seems to be looking for a technical shortcut, forgetting that people have been filming fire for... oh, about 100 years. Even in the last 20 years, it's been possible to capture fire and still get lots of detail in it without clipping.

I've told clients before, "at the least, I feel like there's about a dozen different colors going on in a flame -- provided it's been properly exposed." If all you see is clipped whites and a little yellow, it ain't right. And you don't need dynamic range expansion to get it right. I just saw a fire shot on an episode of The Following where I thought, "wow, the fire looks just right on this." So trust me, it can be done, assuming great lighting, great exposures, and great post work. There are no shortcuts. I think both Epic and Alexa already have enough range as-is to do it normally if you work at it.
 
Marc Wielage said:
No, the real solution is getting somebody who knows how to light the scene and expose it for maximum results. Everybody here seems to be looking for a technical shortcut, forgetting that people have been filming fire for... oh, about 100 years. Even in the last 20 years, it's been possible to capture fire and still get lots of detail in it without clipping.
Click to expand...

LOL.

Your comments about exposure are correct. And this is why this was just from a test shoot.

But you seem to have missed the title of the thread.
It was about how HDRx can't cope very well with the motion of the flames.
 
You must log in or register to reply here.
Back
Top