Panasonic VCR Settings

[Traderbam]

is this better to keep on or off? It looks much less noisy when ON but not sure if I should be using a script to sort it out in post

Capture is much better with off mode. You switched to on mode at frame 39 in your sample, here a post-processing where the first 38 frames have been denoised, and from frame 39 to the end the video is untouched:


[Attachment 83798 - Click to enlarge]

Judge yourself

video_org=AviSource("test1-00.02.01.944-00.02.04.866.avi")

# trimming 1
	trim_start_1=0
	trim_end_1=38
video_org_trim_1=video_org.trim(trim_start_1,trim_end_1)

# trimming 2
	trim_start_2=39
	trim_end_2=0
video_org_trim_2=video_org.trim(trim_start_2,trim_end_2)

### convert to YV16
video_org_trim_1_yv16=video_org_trim_1.convertToYV16()

### denoising
denoised_yv16=video_org_trim_1_yv16.TemporalDegrain2(degrainTR=3)

### convert to YUY2
denoised=denoised_yv16.ConvertToYUY2()

video_final=denoised++video_org_trim_2

return(video_final)

Thanks lollo, difference is incredible. I need to try a comparison now between SOFT and AUTO modes with DNR off. When I tried it before I couldn't see anything as I toggled but will try a capture to verify. I don't know how AUTO is determined and nothing in the manual makes it clear.

[Traderbam]

is this better to keep on or off? It looks much less noisy when ON but not sure if I should be using a script to sort it out in post

Capture is much better with off mode. You switched to on mode at frame 39 in your sample, here a post-processing where the first 38 frames have been denoised, and from frame 39 to the end the video is untouched:


[Attachment 83798 - Click to enlarge]

Judge yourself

video_org=AviSource("test1-00.02.01.944-00.02.04.866.avi")

# trimming 1
	trim_start_1=0
	trim_end_1=38
video_org_trim_1=video_org.trim(trim_start_1,trim_end_1)

# trimming 2
	trim_start_2=39
	trim_end_2=0
video_org_trim_2=video_org.trim(trim_start_2,trim_end_2)

### convert to YV16
video_org_trim_1_yv16=video_org_trim_1.convertToYV16()

### denoising
denoised_yv16=video_org_trim_1_yv16.TemporalDegrain2(degrainTR=3)

### convert to YUY2
denoised=denoised_yv16.ConvertToYUY2()

video_final=denoised++video_org_trim_2

return(video_final)

Lollo, is there any issue running TD2 alongside QTGMC in a script? I usually use something like "faster" preset in QTGMC. Is there a risk of double-denoising and how can I avoid that?

[lollo]

QTGMC denoises by itself and is not easy to turn completely off this filtering, being essential part of the nice outcomes we all appreciate.

The preset "fast" is not recommended when QTGMC is used "alone", and useless for the purpose to avoid over processing when applying additional denoising step.

I generally use something like:

QTGMC(preset="slow", matchpreset="slow", matchpreset2="slow", sourcematch=3, tr1=2, tr2=1, NoiseTR=2, sharpness=0.1)
TemporalDegrain2(degrainTR=3)

but experiment youself with your material, eventually reducing the temporal radius (degrainTR) of TD2.

Some useful old discussion here:
https://forum.videohelp.com/threads/404164-Why-is-QTGMC-so-destructive-and-why-do-so-m...it#post2641895
https://forum.videohelp.com/threads/410894-Avoid-noise-reduction-with-QTGMC

[Traderbam]

That's good to know, I've obviously been using some sample scripts incorrectly.

If I start with these settings, is it adequate to reduce the degrainTR settings from 3 to a lower value, or do we need to start throwing grainLevel,postFFT and postSigma settings into the mix? I much prefer to keep it as simple as possible.

[lollo]

In general, TemporalDegrain2 works quite well with degrainTR=3, so start with that, and experiment lower values.

The post-processing options in TD2 are there to provide additional denoise, which is rarely required. For very bad sources with residual noise after TD2 you can try a spatial denoiser like KNLMeansCL, enabling it inside TD2 (with postFFT and postSigma) or in stand-alone mode (preferred, so you have full control of all parameters).

But, once more, I adhere to the concept "less is better", so be careful to do not over process your videos.

P.S.: remeber that QTGMC is only needed for interlaced sources.

[Traderbam]

Yes all my sources are interlaced.

I've added a few other samples from a recent tape. Some sections of the tape are much better than others so I'm assuming this was recorded to VHS from a camcorder.

The TG2 noise reduction helps a bit on this but it seems the source is overly blurry. I'm sure sure how to describe this technical but if you had any assistance on these clips that could be useful to understand how I can improve. It just seems overly soft to begin with.

[lollo]

A quick attempt on my side. The source is not the best, but I see some improvement with a basic processing.

Levels video 1 OK:



Levels video 2 OK:



Processing video 1:

video (cutted for upload limits): a_cutted.avi

comparison by side:


comparison with slider: https://imgsli.com/MzI0MzUw (a bit unfair, interlaced vs progressive)

Processing video 2:

video: b.avi

comparison by side:


comparison with slider: https://imgsli.com/MzI0MzUx (a bit unfair, interlaced vs progressive)

Filtering video 1:

video_org=AviSource("VHS-TAPE1-Baling-4-00.09.11.019-00.09.16.842-seg1.avi")

# cropping 
	crop_left=16	# | rimozione esatta delle bande nere sinistra, sopra, destra e del disturbo sotto	
	crop_top=8	# | 720-(16+10)x576-(8+4)=694x564
	crop_right=10
	crop_bottom=4
video_org_crop=video_org.crop(crop_left,crop_top,-crop_right,-crop_bottom)

### de-interlacing
deinterlaced=video_org_crop.AssumeTFF().QTGMC(preset="slow", matchpreset="slow", matchpreset2="slow", sourcematch=3, tr1=2, tr2=1, NoiseTR=2, sharpness=0.1)

### convert to YV16
deinterlaced_yv16=deinterlaced.convertToYV16()
### denoising
denoised_yv16=deinterlaced_yv16.TemporalDegrain2(degrainTR=3)
### convert to YUY2
denoised=denoised_yv16.convertToYUY2()

### convert to YV12
video_org_yv12=denoised.convertToYV12()
### sharpening
sharpened_yv12=video_org_yv12.LSFmod(defaults="slow")
### convert to YUY2 with chroma from YUY2 color space
sharpened=sharpened_yv12.convertToYUY2().MergeChroma(denoised)
### add borders
video_rest=sharpened.addborders((crop_left+crop_right)/2-1,(crop_top+crop_bottom)/2,(crop_left+crop_right)/2+1,(crop_top+crop_bottom)/2)

return(video_rest)

Filtering video 2 (just crop top changed):

# cropping 
	crop_left=16	# | rimozione esatta delle bande nere sinistra, sopra, destra e del disturbo sotto	
	crop_top=10	# | 720-(16+10)x576-(10+4)=694x562
	crop_right=10
	crop_bottom=4

[Traderbam]

A quick attempt on my side. The source is not the best, but I see some improvement with a basic processing.

Levels video 1 OK:


[Attachment 83944 - Click to enlarge]

Levels video 2 OK:


[Attachment 83945 - Click to enlarge]

Processing video 1:

video (cutted for upload limits):
[Attachment 83948 - Click to enlarge]

comparison by side:

[Attachment 83946 - Click to enlarge]

comparison with slider: https://imgsli.com/MzI0MzUw (a bit unfair, interlaced vs progressive)

Processing video 2:

video:
[Attachment 83949 - Click to enlarge]

comparison by side:

[Attachment 83947 - Click to enlarge]

comparison with slider: https://imgsli.com/MzI0MzUx (a bit unfair, interlaced vs progressive)

Filtering video 1:

video_org=AviSource("VHS-TAPE1-Baling-4-00.09.11.019-00.09.16.842-seg1.avi")

# cropping 
	crop_left=16	# | rimozione esatta delle bande nere sinistra, sopra, destra e del disturbo sotto	
	crop_top=8	# | 720-(16+10)x576-(8+4)=694x564
	crop_right=10
	crop_bottom=4
video_org_crop=video_org.crop(crop_left,crop_top,-crop_right,-crop_bottom)

### de-interlacing
deinterlaced=video_org_crop.AssumeTFF().QTGMC(preset="slow", matchpreset="slow", matchpreset2="slow", sourcematch=3, tr1=2, tr2=1, NoiseTR=2, sharpness=0.1)

### convert to YV16
deinterlaced_yv16=deinterlaced.convertToYV16()
### denoising
denoised_yv16=deinterlaced_yv16.TemporalDegrain2(degrainTR=3)
### convert to YUY2
denoised=denoised_yv16.convertToYUY2()

### convert to YV12
video_org_yv12=denoised.convertToYV12()
### sharpening
sharpened_yv12=video_org_yv12.LSFmod(defaults="slow")
### convert to YUY2 with chroma from YUY2 color space
sharpened=sharpened_yv12.convertToYUY2().MergeChroma(denoised)
### add borders
video_rest=sharpened.addborders((crop_left+crop_right)/2-1,(crop_top+crop_bottom)/2,(crop_left+crop_right)/2+1,(crop_top+crop_bottom)/2)

return(video_rest)

Filtering video 2 (just crop top changed):

# cropping 
	crop_left=16	# | rimozione esatta delle bande nere sinistra, sopra, destra e del disturbo sotto	
	crop_top=10	# | 720-(16+10)x576-(10+4)=694x562
	crop_right=10
	crop_bottom=4

Thanks for that. I was using very similar - TG2 and LSFMod albeit the fast mode. I'm using Staxrip so the conversions and mergechroma isn't something I specify (assume it happens behind the scenes).

Is the low quality source to be expected if this was recorded from some other device e.g. camcorder to VHS?

[lollo]

Is the low quality source to be expected if this was recorded from some other device e.g. camcorder to VHS?

Yes. Every analog copy degrades the quality. But other aspects may be relevant (i.e. quality of the recorder, quality of the player, signal path, etc.)

I'm using Staxrip so the conversions and mergechroma isn't something I specify (assume it happens behind the scenes).

MergeChroma and ColorSpace conversions are there just because my setup. Actually the first avoid sharpening the chroma, which may be useful in some cases (anyhow, by default the best sharpener do not act on chroma).

You do not need any useless gui, for encoding just use:

ffmpeg.exe -i <input>.avs -c:v libx264 -crf 17 -preset slow -aspect 4:3 -c:a aac -b:a 128k <output>.mp4

[Traderbam]

I'll update this thread with some more test caps shortly, but perhaps a dumb questions. I'm test capturing some tapes on both a Philips (JVC clone) and a Panasonic HS860. Over the course of a 3 hour tape the compression ratio that I get in VDUB is different depending on the VCRs. Philips are about 75GB for 3 hours capture, while the Panasonic is closer to 90GB on the same tape. Is this expected? I have the 3D DNR off on the Panasonic which generates a lot more noise but I'd have assumped the capture throughput was the same bandwidth/rate regardless?

[rgr]

In general, TemporalDegrain2 works quite well with degrainTR=3

When I tested TemporalDegrain2, any value above 1 effectively destroyed fine details.

[lollo]

Your test, my results are different.

edit P.S.: you swear for Neat Video and say that TD2 destroy details. Unbelievable!

[rgr]

Your test, my results are different.

edit P.S.: you swear for Neat Video and say that TD2 destroy details. Unbelievable!

It's unbelievable that any denoiser can destroy details...

PS. More test...

[lollo]

It's unbelievable that any denoiser can destroy details...

Unfortunately is a consequence. Is not easy to distinguish details from noise.

[Traderbam]

Lollo, in your QTGMC settings...

QTGMC(preset="slow", matchpreset="slow", matchpreset2="slow", sourcematch=3, tr1=2, tr2=1, NoiseTR=2, sharpness=0.1)

I take it the purpose here is to preserve as much of the original detail as part of the deinterlace so that the TG2 noise removal is working with as original a source as possible? If for a given clip I wasn't as concerned on quality, and preferred speed, would I be better placed to use QTGMC "fast" without TG2?

TG2 and LSFMod slow down processing a lot, so I'm trying to find a "lightweight" option for QTGMC that will give me a compromise.

I assume "fast" default preset will remove more noise, and not reduce the sharpening (like you have with 0.1 setting).

Thanks

[lollo]

If for a given clip I wasn't as concerned on quality, and preferred speed, would I be better placed to use QTGMC "fast" without TG2?

That setting on QTGMC is to reduce its denoise (not easy) and then specifically use TD2.

The bottleneck in term of speed is TD2, especially with a high temporal radius (i.e. degrainTR=3, default), not QTGMC.

If you are not concerned about quality, remove TD2 from the processing and use QTGMC with one of the basic preset, to keep its denoising capability. For example QTGMC(preset="slow") or QTGMC(preset="fast"); the second provide less accurate fine tuning for moving objects and less denoise, but it can be adequate according to your source.

As always, there is not a generic rule, each source is different.

[Traderbam]

If for a given clip I wasn't as concerned on quality, and preferred speed, would I be better placed to use QTGMC "fast" without TG2?

That setting on QTGMC is to reduce its denoise (not easy) and then specifically use TD2.

The bottleneck in term of speed is TD2, especially with a high temporal radius (i.e. degrainTR=3, default), not QTGMC.

If you are not concerned about quality, remove TD2 from the processing and use QTGMC with one of the basic preset, to keep its denoising capability. For example QTGMC(preset="slow") or QTGMC(preset="fast"); the second provide less accurate fine tuning for moving objects and less denoise, but it can be adequate according to your source.

As always, there is not a generic rule, each source is different.

Helpful, thanks. Most of my clips will end up on youtube so it's probably not worth the effort for TG2 and LSFMod only for YT to kill the quality anyway. I do try to resize to 1440x1080 to hopefully get better bitrate but not convinced.

[lollo]

To reduce Youtube encoding impact you can upload a lossless sample (but the advantage on the final result compared to uploading a h264 compressed with low CRF is marginal https://www.digitalfaq.com/forum/video-capture/11510-parts-build-pc-2.html#post79818) and upscale to HD 1440x1080 (this makes a difference)

[Alwyn]

I've looked closely at a few of my recent uploads and I think this YT destruction idea is way overblown. If your denoising improves your video on your computer, it'll improve your YT video too.

Anyway, if you make your vertical res 1440 or above, you'll get VP9 encoding.

[Traderbam]

To reduce Youtube encoding impact you can upload a lossless sample (but the advantage on the final result compared to uploading a h264 compressed with low CRF is marginal https://www.digitalfaq.com/forum/video-capture/11510-parts-build-pc-2.html#post79818) and upscale to HD 1440x1080 (this makes a difference)

My workflow is to deinterlace from huffyuv avi to ProRes, import into Resolve for cutting, watermarking, logos, overlays etc, then usually render to H265 (MOV) for upload with the "best" profile in Resolve settings. This usually generates file with something like 25000kbps bitrate.

Does YT prefer H264 in mp4 container and is there any quality difference if I were to use H265 instead?

[lollo]

I've looked closely at a few of my recent uploads and I think this YT destruction idea is way overblown. If your denoising improves your video on your computer, it'll improve your YT video too.

I agree

Does YT prefer H264 in mp4 container and is there any quality difference if I were to use H265 instead?

I have no experimented it. But given that H265 for HD resolution gives the same quality at lower bitrate (or higher quality at same bitrate), and that we are talking about very low CRF and upload to youtube, probably it does not make a difference.

[Traderbam]

I use the native encoding option in Resolve and it seems to be based on average bitrate. There is an IntelQSV option which allows me to choose between Intelligent Constant quality, constant bitrate, variable bitrate, constant QP - no idea what these options are so I stick to the easy "native" encoder with "best".

[lollo]

Yes, if upload time and size are not a problem, rendering at maximum quality place you on the right side.

[Traderbam]

Sounds good. The other open question bugging me is how to properly handle the black borders. My OCD tendencies seem to allow this to annoy me.

Capturing at 720x576, I crop 8 off both sides before a resize. On some of the Sony camcorder captures there was a purple bar down the right side which means I need to crop more than 16, then add borders to 'refill' the gap.

How do I go about tacking the top and bottom borders without messing up things and destroying the benefit of having a high fidelity source file? I'm assuming I cannot just crop to something like 680 x 540 without consequence. Are there defined ranges I can work with?

[lollo]

If you define the PAR instead of the DAR you can crop whatever you want and keep the right proportions. Obviously the player must be able to deal with a PAR definition and resize the video accordingly.

If you want to be on the safe side, just mask the defective borders, rebuilding the 720x576 frame prior to upscaling. But you'll end up with black borders on each side. For me is not a problem, for somebody is not nice.

[Traderbam]

My preference is to not have any but it's not mandatory, more like a nice to have. Assuming I am doing the 1440x1080 resize anyway, can I just crop as I need, and everything will be 'fixed' as part of the resize operation, or is this thing going to destroy something?

[Alwyn]

Borders show up on YT like sore thumbs. They are almost as bad as still-visible head-switching noise!

I just crop off in 4:3 proportion. Ignoring the arguable 2x8, if you take 10 off the right for the green line, take 8(ish) total from top and bottom. Do it in even numbers. Then encode to 1440x1080 (or 1920x1440 ).

BTW, re my 1440 comment, you don't need to increase the bitrate when you do that. You'll still get VP9 at any bitrate.

[Traderbam]

Ok, so if I need to crop width to something like 680, I can crop height to 510 to preserve the 4:3 as given by a calculator like this one https://www.aspectratiocalculator.com/4-3.html?

[Alwyn]

Yessir. 40 off the width, 30 off the vertical; 4:3.

I'll qualify my cropping guidelines: if you're taking that much off, then you may end up removing some stuff you want to keep. In that case, you'll need to keep some bars.

[Traderbam]

Yessir. 40 off the width, 30 off the vertical; 4:3.

I'll qualify my cropping guidelines: if you're taking that much off, then you may end up removing some stuff you want to keep. In that case, you'll need to keep some bars.

720 x 576 is my starting point which isn't 4:3 - do I need to resize to 720x540 first? I'm in PAL land.