4k to 1080

[MaxKitaev]

Hello,

i use the new Panny G7 for couple of weeks and wanted to know the best way to downscale 4k to 1080 with quality/colour gain (422, 444 whatever works) mentioned so many times on web.

Please advise the best way to convert.

[Julian]

I just make a 1080p project and scale the footage to 50% width and height (or change it if you want to crop, recompose or zoom). In Premiere Pro this looks absolutely fine without scaling artefacts. I don't believe there's a noticeable difference in doing it in fancy ways.

If anything can prove me wrong I'd be happy to know though.

[dhessel]

Yep just downscale in whatever app you use. The whole 4K to HD 10bit theory was incorrect so you will gain nothing by tracking down the specialty apps that were made for this.

[tupp]

The whole 4K to HD 10bit theory was incorrect so you will gain nothing by tracking down the specialty apps that were made for this.

​No.  The "4k to HD" theory is correct.  One certainly can sacrifice resolution for a gain in bit depth.

 

However, such a conversion cannot increase the color depth -- one can never increase the color depth of an image without introducing something artificial.

[TheRenaissanceMan]

​No.  The "4k to HD" theory is correct.  One certainly can sacrifice resolution for a gain in bit depth.

 

However, such a conversion cannot increase the color depth -- one can never increase the color depth of an image without introducing something artificial.

​The theory that one can get 10-bit HD from 8-bit 4K has been disproven. The bit depth increase is paltry--something like 8.67 bits instead of 8. 

If you want more robust footage to grade, hook up an external recorder. Otherwise, just downscale the 4K normally and revel in the detail. 

[tupp]

​The theory that one can get 10-bit HD from 8-bit 4K has been disproven. The bit depth increase is paltry--something like 8.67 bits instead of 8. 
If you want more robust footage to grade, hook up an external recorder. Otherwise, just downscale the 4K normally and revel in the detail. 

​That was discussed here.  As I recall, the "8.67" figure came from a scenario in which the original Bayer image was chroma subsampled.  Using the same math that yielded the "effective 8.67 bit depth" would give an "effective bit depth" of less than "8" to the original subsampled image.

 

For simplification, it is best to consider RGB scenarios with no chroma subsampling.  Going from UHD to HD crams 4 pixels into one.  Thus, in an efficient summed conversion, a 4X increase the bit depth is possible, yielding a bit depth of 10.  Such conversions have been done and are being done.

 

Again, with these conversions, we are merely swapping resolution for bit depth -- we are not increasing color depth.  The color depth of an image cannot be increased without adding something artificial.

 

Nevertheless, such bit depth increases also work with chroma subsampled images.  It's just that converting a subsampled image with less "effective bit depth" will yield a reduced "effective bit depth" in the final  image.

[Guest Ebrahim Saadawi]

There are many ''algorithms'' for downsampling, and the best ones (cubic & Lanczos) are used in your NLE when you put 4K footage on a 1080p timelime or when you render 4K timeline to 1080p. Don't worry there isn't a better way. 

[mercer]

There are many ''algorithms'' for downsampling, and the best ones (cubic & Lanczos) are used in your NLE when you put 4K footage on a 1080p timelime or when you render 4K timeline to 1080p. Don't worry there isn't a better way. 

Why wouldn't you transcode before you bring it into your NLE?

[Guest Ebrahim Saadawi]

Mainly because the time lost transcoding and the inability to crop/reframe in the edit. I suggest not downsampling before ingesting, work with the 4K files.

[Sekhar]

If you scale in Premiere Pro, be sure to check the "Use Maximum Render Quality" in the export dialog, apparently that scales at a higher quality. There is some debate on whether or not a CUDA render option automatically selects this option, but it's better to check it yourself to be sure.

[tupp]

There are many ''algorithms'' for downsampling, and the best ones (cubic & Lanczos) are used in your NLE when you put 4K footage on a 1080p timelime or when you render 4K timeline to 1080p. Don't worry there isn't a better way. 

​If you reduce resolution without increasing bit depth, you are just throwing away color depth information.

 

Do you think that Technicolor, FotoKem  and DeLuxe are using NLEs to down-convert studio features, or do you think that they try to retain as much color depth info as possible?