Is converting dv 4:2:0 footage to 4:4:4 worth it for restoration?

[taigi]

I’m working on restoring old DV footage (8-bit, YUV 4:2:0) that needs a lot of correction, noise reduction, and etc.
I’ve read that converting early on to a higher sampling format like YUV 4:4:4 can help preserve quality.

For those with experience in video restoration:

Is it really worth converting to 4:4:4 as an intermediate for this kind of work, or is 4:2:2 enough? Does converting to 4:4:4 actually help with quality when the original footage is only 4:2:0? At what stage in the workflow do you recommend doing this conversion, if at all?

Any advice would be greatly appreciated. Thanks.

[Sharc]

I’m working on restoring old DV footage (8-bit, YUV 4:2:0) that needs a lot of correction, noise reduction, and etc.
I’ve read that converting early on to a higher sampling format like YUV 4:4:4 can help preserve quality.

For those with experience in video restoration:

Is it really worth converting to 4:4:4 as an intermediate for this kind of work, or is 4:2:2 enough? Does converting to 4:4:4 actually help with quality when the original footage is only 4:2:0? At what stage in the workflow do you recommend doing this conversion, if at all?

Any advice would be greatly appreciated. Thanks.

No. Upsampling to 4:4:4 is a waste of storage space in your case, IMO.

Your PAL DV footage is interlaced 4:2:0. Most filters work perfectly in YV12 (4:2:0) colorspace, so no need to upsample the chroma from this point of view. More important is the deinterlacing, as many filters have no interlace-aware option and work properly on progressive footage only. If you want to keep it interlaced make sure to apply the filters correctly (on fields or grouped fields, depending on the filter).

I assume you transferred the PAL DV to PC digitally (via firewire) as 4:2:0. If for some reason you have to capture it analog, capture it lossless 4:2:2 interlaced, revisit the levels and proceed from there with post processing. In general, avoid conversions (especially any YUV<->RGB) unless a filter requests a particular format. If you do the processing in 4:2:2 you may want to convert it to 4:2:0 at the end (for distribution), for best player compatibility.

[themaster1]

Educated users of avisynth may disagree on the matter converting to yv16 or yv24 (4.2.2 and 4.4.4 respectively) is the right formulae to get outstanding results for noise-less videos and colorful / vibrant with yv16/24 aware filters.
The downside is take more horsepower

[Selur]

Not sure whether I would count myself as an 'Educated users of avisynth' nowadays (I rarely use it), but I would like to see an example where one:

• starts with 4:1:1 or 4:2:0 content, upsamples to 4:4:4 or 4:2:2, then
• applies filtering, that also would have worked with the original color sampling, and then
• downsamples back to 4:2:0 or 4:1:1 for the output

and one can really see an improvement in the output.

Sure, when aiming for higher sampled output and filter accordingly, often upsampling earlier in the filter chain can make sense.
Increasing the bit depth / precision for filtering might help in some rare cases, but I doubt that upsampling the color sampling makes sense in the above scenario.

The downside is take more horsepower

That is usually negligible when applying CPU intensive filtering anyway and are not trying to use minimal filtering to preserve real time or higher processing speeds.

Cu Selur

[taigi]

in some rare cases

Could you please share some specific cases where upsampling to 4:4:4 or increasing bit depth before filtering makes a difference?
I’m currently working on very dark footage from a weird-lit club. Would upsampling beforehand be helpful in this situation?
Thank you!

[dellsam34]

If you are doing color correction and other visual restorations it doesn't hurt to drop the footage in vdub2 and output AVI 4:2:2 into say HuffYUV, But I'm skeptical of noticing a big visual difference even though it makes some technical sense.

[Selur]

I’m currently working on very dark footage from a weird-lit club. Would upsampling beforehand be helpful in this situation?

Highly unlikely. Usually it only might help, if you have high contrast colors to begin with, which should not be the case in a 'very dark footage from a weird-lit club'.

[rgr]

Converting to 10 or 16 bits will yield better results, though that doesn't mean the difference will be particularly noticeable (but that depends on the source and other factors).

It also depends on how many filters are used (rounding accumulator). If you only use one, it's not worth it. However, I've noticed that even a single TemporalDegrain2 handles banding better at 10 bits.

Personally, I'd work in the 10-bit YUV422 range.
And personally I work in YUV444P16