EXR or DPX for my 8-bit Panasonic lumix g7 4k UHD-files

Hello there,

I am filming with a Panasonic lumix G7 camera it records my footage into 4K uhd 8-bit 4:2:0

I want to trim and color correct my clip in Davinci Resolve12 and then export the clip so i can work with it in Fusion.

I have done some testing to find out which export format would be best. I am doubting between EXR or DPX

-The DPX bit depth is: 16bit Int
-The EXR bit depth is: 16bit Float

My question is:
- can i transform my 8 bit 4k uhd footage into a 16 bit ?
-And which would be better 16bit Int or 16bit Float ?

-And which would work better in Fusion, cause the EXR file is about 35% bigger then the DPX file

I hope that somebody can help

Greetings Lenny

because you only have 8bit material as source you want gain any precision or more data if you convert to 16bit int or 16bit float

i thinks its just a matter of personal choice in this case if you go with dpx or exr.
both work very well in fusion and can be read and written

Hey,

Industry convention is 10bit (usually log) DPX for editorial exchange. Half float linear EXR (16bit) for VFX exchange. If you stick to that then you can't go too far wrong. But have a play, do some extreme tests. I only found a benefit with EXR for RAW footage with EXTREME grades in my tests.

because you only have 8bit material as source you want gain any precision or more data if you convert to 16bit int or 16bit float

i thinks its just a matter of personal choice in this case if you go with dpx or exr.
both work very well in fusion and can be read and written

Michael, technically you are right of course, although Stu Maschwitz suggests something smart in this post:

http://prolost.com/blog/2010/1/26/color ... otage.html

If you do a really subtle denoise on your 4:2:0 footage in a higher bitdepth, the blurring will create 'new' colori nformation which will be smoother because of the resampling. I'm pretty sure this will work in Fusion as well. Nils will just have to test whether or not exporting at a higher bit depth from Resolve will result in those fake smoother values.

Nils, how are you shoothing the footage? Log or lineair?

because you only have 8bit material as source you want gain any precision or more data if you convert to 16bit int or 16bit float

i thinks its just a matter of personal choice in this case if you go with dpx or exr.
both work very well in fusion and can be read and written

Michael, technically you are right of course, although Stu Maschwitz suggests something smart in this post:

http://prolost.com/blog/2010/1/26/color ... otage.html

If you do a really subtle denoise on your 4:2:0 footage in a higher bitdepth, the blurring will create 'new' colori nformation which will be smoother because of the resampling. I'm pretty sure this will work in Fusion as well. Nils will just have to test whether or not exporting at a higher bit depth from Resolve will result in those fake smoother values.

Nils, how are you shoothing the footage? Log or lineair?

Thanx guys for all the comments and tips.

What is the difference between Log or Lineair?
I am sorry to be a newbee But you gotta start somewhere

My footage is always green screen footage. So if i do this denoising and upscaling to a higher bit depth in Resolve. Would it be better then to do the keying in Resolve --> before the upscaling. Or do the keying after the upscaling --> in Fusion ?

Or could i also upscale to a higher bit depth in Resolve and then reload the exported file into resolve again and then do the keying?

In other words would the upscaling to a higher bit depth, affect the result of my keying?

Thanx in front
Greetz Lenny

Up the bitdepth and denoise in Resolve, bring the shot out as dpx files and then proceed with the keying in Fusion. This will result in a decent key, probably a nicer one than doing it before increasing the bit depth.

As far as I know the G7 doesn't shoot log, think that's reserved for the GH4.

For a bit of info about the difference between log and linear, have a quick look at this, it'll explain better than I could;
http://www.qvolabs.com/Digital_Images_C ... inear.html

Up the bitdepth and denoise in Resolve, bring the shot out as dpx files and then proceed with the keying in Fusion. This will result in a decent key, probably a nicer one than doing it before increasing the bit depth.

As far as I know the G7 doesn't shoot log, think that's reserved for the GH4.

For a bit of info about the difference between log and linear, have a quick look at this, it'll explain better than I could;
http://www.qvolabs.com/Digital_Images_C ... inear.html

Thanx sam for the input and information. I will do the keying in fusion. I am glad that the workflow is better like this. Cause i am more familiar with the fusion keying techniques then with the resolve ones.

I also will dive in the article, but Maybe you can answer this question?

Is Log: 10-bit 4:2:2 ?
and Linear: 8-bit 4:2:0 ?

Or is this a different thing?

Greetz lenny

totally different things....

Log is a logarithmic data 'curve' that stores data in a non uniform way....usually more bits dedicated towards the highlights. It will not look proper on a monitor without a LUT.

Linear is an even distribution of data. It also will not look proper without a LUT.

Log can squeeze more usable data into a smaller file(usually) and a 10bit log is roughly equal to a 12bit linear in terms of camera recording.

Both are post production formats and made to store a lot more visual information vs a 709/srgb file which stores data in the limited color space and bit depth of the file...but is ready for immediate viewing


10 bit or 8 bit can be either LOG or linear.....that is just the amount of data assigned to the specific file.However, linear files are commonly 12 bit and up....with a lot of formats being 16 or 32 bit.

8 bit has 256 shades per color. 10 bit has 1024. 16 bits has over 32,000......and 32 bits a lot more

4:2:2 and 4:2:0 are codec color sub sampling patterns based on 4 pixel sample.

4 represents green, which is always full sample. The next numbers are the blue and red channels which in 4:2:2 are discarding half the red and blue data (averaging)

4:2:0 has a weird naming, but samples it's 4 pixels in a rectangle, rather than a line.....it fully samples green, and discards 75% of the red and blue channels. It is the same color information as 4:1:1 which also samples in a line vs a quadrant.

There are also a lot of 4:4:4 color formats....which are full sampled in all color channels. If you see 4:4:4:4 then there is an alpha channel as well (always fully sampled)

http://www.dvxuser.com/articles/colorspace/

Then you have RGB and YUV...which I will let you discover

I've never seen a codec with 4:2:2 or 4:2:0 RGB subsampling. Which ones use that?

I've never seen a codec with 4:2:2 or 4:2:0 RGB subsampling. Which ones use that?

No RGB subsampling codecs (AFAIK).....I used RGB interchangeably for describing YUV, YCbCr, YPbPr as to not confuse the OP