Click here to go to the first RED TEAM post in this thread.   Thread: Helium IPP2 Highlight Test - Huge improvement

Reply to Thread
Page 1 of 15 1234511 ... LastLast
Results 1 to 10 of 142
  1. #1 Helium IPP2 Highlight Test - Huge improvement 
    Senior Member
    Join Date
    Dec 2006
    Location
    Toronto, Canada
    Posts
    819
    A few months ago NoFilmSchool.com published a "ARRI Alexa Mini vs. RED Epic-W Helium" shootout test. They concluded:

    "The Alexa has at least a stop and half more dynamic range in the highlights when compared to the Helium."

    That was a surprise, so I wanted to test Red's new IPP2 colour science to see how it handled extreme highlights. I discovered that an additional 1-1.5 stops could be extracted from the highlights without the hard clipping seen in the original test. IPP2 is still in beta, but the early results demonstrate much greater highlight retention, more comparable to Alexa. These grades are dealing with extreme highlights (4-6 stops overexposed) working in ISOs from 15-50 iso!!! The images are certainly NOT showing either sensor in ideal circumstances, but it's always nice to know how things clip. My rough grades were done in Redcine-X Pro, beta build 50.42644. Secondary colour was done in Autodesk Smoke. The grades are not an attempt to match Alexa. They are just a quick exploration of the highlight data hiding in each R3D.

    Other tests have revealed a similar noise floor between Helium and Alexa, yielding comparable shadow info and low-light capabilities. And now it looks like IPP2 is closing the highlight & rolloff gap. Kudos to Graeme and Red team for their stellar work on IPP2.

    I also tested the original Alex Log-C DPX files, and was able to squeeze a tiny bit more info out of them, but nothing too dramatic. The results are posted here:

    A video DEMO of my test can be seen here:
    https://vimeo.com/217270940

    Graded 4K Tiffs can be found here:
    dropbox.com/sh/6ze4gdahoepp4zl/AABy68TEtTlYhsDqsbuj2R-ta?dl=0

    Special thanks to No Film School, Charlie Anderson, Alex Chinnici, Justin Derry, and Oren Soffer for providing R3Ds and DPX for testing purposes. Their original video can be seen here:
    vimeo.com/202842588.

    Their analysis is here:
    nofilmschool.com/2017/02/shootout-arri-alexa-mini-vs-red-epic-w-helium
    Attached Images
    Reply With Quote  
     

  2. #2  
    Junior Member
    Join Date
    Oct 2009
    Posts
    24
    That's fantastic!

    Could I ask which IPP2 contrast and rolloff settings you used?
    Reply With Quote  
     

  3. #3  
    Senior Member
    Join Date
    Dec 2006
    Location
    Toronto, Canada
    Posts
    819
    I believe I used low contrast and very soft. ISO was set to 50, and exposure adjust between -0.5 and -2. Also due to the warm key light, the best highlight rolloff was @ col temp 1800 Kelvin. I snapped things up and tweaked in Autodesk Smoke, which is my usual workflow. I'm a complete IPP2 novice, so I may not be doing things the best way.
    Reply With Quote  
     

  4. #4  
    Senior Member
    Join Date
    Dec 2014
    Location
    Sarajevo
    Posts
    938
    Un...believable!

    You sure You used the same, the "5 stops over", clip?
    Reply With Quote  
     

  5.   Click here to go to the next RED TEAM post in this thread.
  #5  
    Quote Originally Posted by Šabović Adis View Post
    Un...believable!

    You sure You used the same, the "5 stops over", clip?
    Yes he did! I was able to duplicate what Alex had done after speaking to him on this the other night. New IPP2 highlight extension algorithm helps a lot, but then you have to bring it back with the grade, but it's all quite doable.

    Graeme
    www.red.com - 8k Digital Cinema Camera
    Science enables stories. Stories drive science
    IPP2, Image Processing, Colour Science and Demosaic Algorithms
    Reply With Quote  
     

  6. #6  
    Senior Member
    Join Date
    Dec 2014
    Location
    Sarajevo
    Posts
    938
    Quote Originally Posted by Graeme Nattress View Post
    Yes he did! I was able to duplicate what Alex had done after speaking to him on this the other night. New IPP2 highlight extension algorithm helps a lot, but then you have to bring it back with the grade, but it's all quite doable.

    Graeme
    Butbut... we're talking 1.5 stops here!
    What did You do, G?

    Reply With Quote  
     

  7. #7  
    Senior Member Thai Christen's Avatar
    Join Date
    Feb 2013
    Location
    Switzerland
    Posts
    493
    Quote Originally Posted by Graeme Nattress View Post
    Yes he did! I was able to duplicate what Alex had done after speaking to him on this the other night. New IPP2 highlight extension algorithm helps a lot, but then you have to bring it back with the grade, but it's all quite doable.

    Graeme
    Fantastic Job Graeme!
    When you said you have to bring it back with the grade, you mean Non-Helium sensor, right? Because with Epic-W I clearly see the highlight improvement and rolloff with IPP2 in camera. Or I am missing something?
    Reply With Quote  
     

  8. #8  
    Senior Member
    Join Date
    Dec 2006
    Location
    Toronto, Canada
    Posts
    819
    Quote Originally Posted by Thai Christen View Post
    Fantastic Job Graeme!
    When you said you have to bring it back with the grade, you mean Non-Helium sensor, right? Because with Epic-W I clearly see the highlight improvement and rolloff with IPP2 in camera. Or I am missing something?
    On this particular shot the smoothest highlight rolloff was visible at 1800 Kelvin, meaning the shot needed to warmed-up after "the primary grade". I chose to do this in Smoke (due to my quirky habits) but any decent grading platform would have worked.

    And yes, it's the exact same shot (same frame) as the nofilmschool test. You can see this in the Vimeo DEMO and TIFFs linked above.

    I didn't shoot the material, so I can't definitively say 1.5 stops, but the results show the IPP2 face-clipping starts between "5-6 stops over".... vs "4 stops over" using the legacy color pipeline. Also the clipping has a much smoother rolloff. It's pretty amazing to see half the dude's reappear.
    Reply With Quote  
     

  9. #9  
    Senior Member Thai Christen's Avatar
    Join Date
    Feb 2013
    Location
    Switzerland
    Posts
    493
    Quote Originally Posted by Alex Boothby View Post
    On this particular shot the smoothest highlight rolloff was visible at 1800 Kelvin, meaning the shot needed to warmed-up after "the primary grade". I chose to do this in Smoke (due to my quirky habits) but any decent grading platform would have worked.

    And yes, it's the exact same shot (same frame) as the nofilmschool test. You can see this in the Vimeo DEMO and TIFFs linked above.

    I didn't shoot the material, so I can't definitively say 1.5 stops, but the results show the face-clipping starts between "5-6 stops over".... vs "4 stops over" using the legacy color pipeline. Also the clipping has a much smoother rolloff. It's pretty amazing to see half the dude's reappear.
    Thanks Alex.
    I don't know about other Epic-W shooters, but it is very noticeable. With IPP2 (medium and soft-very soft) highlight roll off much smoother. A couple of weeks ago I did a quick non scientific test with EpicW and EpicD (both camera at ISO 800, 5600 Kelvin, comparing file in RCX beta). I would even say, EpicW with IPP2 rolloff is a tad better than EpicD. DR about the same. Maybe someone else with better knowledge and methodology can confirm.
    Last edited by Thai Christen; 05-13-2017 at 09:18 AM. Reason: Add additional info
    Reply With Quote  
     

  10. #10  
    Senior Member Anton Shavlik's Avatar
    Join Date
    Aug 2014
    Location
    Hollywood
    Posts
    527
    Any chance this roll off could be accessible more generally in the color science? i.e. Developing at 25iso 1800k vs correct settings for the scene and getting the same highlight retention in either case. Or is this a technique that needs some human decision making to make look correct?

    edit: testing for myself it seems that clip's results are so impressive because the scene was actually lit near 1800k. Anything recovered this way is brought back monochrome. The key is to white balance on the pixel on the edge of your clipped area, so that this pixel has no hue, then it will roll off into the recovered monochrome highlights perfectly. Then you bring everything back from there and it looks ok. But there will never be subtly in the hues or saturation of the recovered areas.

    I tried this on another of their samples, where a white window curtain was blown out with a daylight HMI. The best roll off retention was at 5800k-12, i.e. white balancing to the white window curtain where it wasn't blown out, iso and roll-off/contrast settings did not have any effect beyond what they normally do.
    Last edited by Anton Shavlik; 05-13-2017 at 09:30 AM.
    Reply With Quote  
     

Posting Permissions
  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts