RAW Video of What Sensor Sees

[maxotics]

Half the things I try with RAW end up no-where--just can't figure it out.  Then one day it clicks.  Last night I was finally able to create a video that shows the bayer pattern recorded in RAW, before debayering.  I hope to do more in the future. 

 

I wish I had seen a video like this early on, so I would have visually grasped the importance of what de-bayering algorithm is used.  Also, would see why one dead sensor pixel is insignificant.

 

https://vimeo.com/79857693

[HurtinMinorKey]

^Fucking awesome. It's the first time I've seen something like this. It's even greener than I thought it would be.

 

I've always questioned the 4:2:2 ratio. I know it's supposed to be based on empirical  measurements of human sensitivity, but it sounds way too general (and way too convenient) to be true.

[mtheory]

What is the significance of this in relation to green screen vs blue screen work? 

[maxotics]

What is the significance of this in relation to green screen vs blue screen work? 

 

In a perfect world, you'd want each pixel to have a perfect green or blue reading.  But they don't.  Only half of them have green values and a quarter blue.  AFAIK, in practical terms, after interpolation, chroma keys works perfectly.  Theoretically, one might reason that green screens should be more accurate.  I don't know. 

 

A take away is that, even in RAW video, each pixel is not completely accurate.

 

In video compression, a lot of color information is being thrown away.  If it's properly exposed, that's usually not a problem.  It's when you need to bring up "green" values from the RAW data that you're in a bind.  It wasn't perfect in the first place and now you've lost much of that.

[gloopglop]

@maxotics, wow, i saw this at work earlier today and it blew my mind

 

SO interesting. thank you

 

may i ask how in the heck you created this video exactly?? i would love to know a bit about the process here  :huh:

[maxotics]

may i ask how in the heck you created this video exactly?? i would love to know a bit about the process here  :huh:

 

1. I open a DNG in Photivo and, under "Camera", set the Demosaicing to "Bayer Pattern".  The default box will say DCB.  I will also set the output to the highest JPG setting at 4:4:4.  

2. I create a batch proceess to save all the DNGs in that folder to jpg (but you can save to TIFF etc.).

3. I then use a DOS/CMD script that calls ImageMagick that crops out a section of each JPG and scales it back up

 

for %%f in (M*.jpg) do (convert %%f -gravity Center -crop 170x90+0+0 +repage -scale 600%% crop170_%%f)

for %%f in (M*.jpg) do (convert %%f -gravity Center -crop 85x45+0+0 +repage -scale 1200%% crop85_%%f)

 

I then create clips of the various JPG series and put together in NLE and render.

 

At some point I hope to do a better version, where it would show something recognizable, like someone's face, in the expanded crop.

[gloopglop]

@maxotics, thank you for the explanation

 

i must say that i find this work both technically fascinating and aesthetically compelling

 

youre definitely onto something here

 

id love to see more of this

 

your ambition to show "something recognizable, like someone's face" is very intriguing

 

tbh if i wasnt such an 'original' asshole i would TOTALLY steal your idea for some kind of art installation  B) 

 

love that 'bayer' image

 

kinda just wanna curl up in that sea of green and sleep forever