Michael Tiemann
Well-known member
Dear Dragonistas,
I decided to collect some data this weekend on the behavior of REDCODE compression. I settled on the following test/evaluation theory:
1. All clips are static shots 3 seconds long. When played in a loop, one can get a really good feeling about how the noise behaves (in a static shot).
2. I used what I believe is my sharpest lens for this test: the Canon EF 24mm TS L stopped down to f5.6
3. The frame rate is 12 fps and the shutter is 360 degrees. Shooting at 12fps makes the noise more apparent than when shot at 24fps. Shooting at 1/12th of a second puts more light on the sensor, which reduces noise (generally)
4. The dark wood floor and the shadows along the sides turn out to be great generators of video noise. The high-contrast fine details in the scene don't tend to break down and show up as noise. Thus, noise typically comes from the shadows.
I shot at ratios of 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 18, 20, and 22:1 (in that order). The metadata in the clips will tell you the compression rate for sure. I shot 4K at a 5:4 aspect ratio, providing enough pixels that one can evaluate noise at full-res vs. scaled-down res.
What I have observed in my cursory evaluation is that when viewed at Full/100%, even under these somewhat tortuous conditions (12fps!), REDCODE 2:1 looks better as a moving image than as a still image and whatever noise is there is very, very small and granular. As the REDCODE compression rate moves up, the relative size of the noise seems to increase, until about REDCODE 10:1, when the relative intensity of the noise (and also its prevalence across the image) also increases. At REDCODE 22:1, the noise is very easily seen coming from the entire floor and from the whole back wall.
Now, one of the reasons that DRAGON is a 6K camera is so that one can capture at a higher resolution than the delivery resolution. When these images are scaled down to Full/70%, visible noise is reduced. And when scaled down to Full/50%, it is reduced further. It is interesting to compare the noise character at REDCODE 2:1 at 100% to REDCODE 4:1 at 70% or REDCODE 8:1 at 50%. It is also interesting to stretch out the comparison a bit, so REDCODE 2:1 at 100% vs. REDCODE 5:1 at 70% vs REDCODE 12:1 at 50% vs. REDCODE 22:1 at 25% (sadly there's no 35% setting in REDCINE X).
Anyway, here's a link to the R3D files at different levels of compression, FWIW: https://www.dropbox.com/sh/bhpx7hx6xavyfae/5WjC0rCsBk
I decided to collect some data this weekend on the behavior of REDCODE compression. I settled on the following test/evaluation theory:
1. All clips are static shots 3 seconds long. When played in a loop, one can get a really good feeling about how the noise behaves (in a static shot).
2. I used what I believe is my sharpest lens for this test: the Canon EF 24mm TS L stopped down to f5.6
3. The frame rate is 12 fps and the shutter is 360 degrees. Shooting at 12fps makes the noise more apparent than when shot at 24fps. Shooting at 1/12th of a second puts more light on the sensor, which reduces noise (generally)
4. The dark wood floor and the shadows along the sides turn out to be great generators of video noise. The high-contrast fine details in the scene don't tend to break down and show up as noise. Thus, noise typically comes from the shadows.
I shot at ratios of 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 18, 20, and 22:1 (in that order). The metadata in the clips will tell you the compression rate for sure. I shot 4K at a 5:4 aspect ratio, providing enough pixels that one can evaluate noise at full-res vs. scaled-down res.
What I have observed in my cursory evaluation is that when viewed at Full/100%, even under these somewhat tortuous conditions (12fps!), REDCODE 2:1 looks better as a moving image than as a still image and whatever noise is there is very, very small and granular. As the REDCODE compression rate moves up, the relative size of the noise seems to increase, until about REDCODE 10:1, when the relative intensity of the noise (and also its prevalence across the image) also increases. At REDCODE 22:1, the noise is very easily seen coming from the entire floor and from the whole back wall.
Now, one of the reasons that DRAGON is a 6K camera is so that one can capture at a higher resolution than the delivery resolution. When these images are scaled down to Full/70%, visible noise is reduced. And when scaled down to Full/50%, it is reduced further. It is interesting to compare the noise character at REDCODE 2:1 at 100% to REDCODE 4:1 at 70% or REDCODE 8:1 at 50%. It is also interesting to stretch out the comparison a bit, so REDCODE 2:1 at 100% vs. REDCODE 5:1 at 70% vs REDCODE 12:1 at 50% vs. REDCODE 22:1 at 25% (sadly there's no 35% setting in REDCINE X).
Anyway, here's a link to the R3D files at different levels of compression, FWIW: https://www.dropbox.com/sh/bhpx7hx6xavyfae/5WjC0rCsBk
