Digital zoom with Resolve vs incamera

[SillyPosition]

I recently started to get into videography, nothing "serious" too much, just family occassions.

Yesterday was my daughter ballet concert, so I took my tripod, put my camera into proper settings and while filming I obviously moved my camera a little bit from side to side on the stage, sometimes zoomed a little (Its a 24-70, but manual ring zoom), etc.

Now in resolve I wanted to zoom in and shift around the frame every now and then, so I used transform markers and applied what's "needed", and its a much better and smooth now.

I understand that zooming in and then exporting to 4k does some sort of upscaling to the video, because the input was 4k to begin with, how severe is it to do so?

Its to the point where if its quite good, I think its much easier shooting wider being perfectly still, and do it all in post, getting a very smooth zoom/movement operation

 

[Joe Lynch]

I understand that zooming in and then exporting to 4k does some sort of upscaling to the video, because the input was 4k to begin with, how severe is it to do so?

Its to the point where if its quite good, I think its much easier shooting wider being perfectly still, and do it all in post, getting a very smooth zoom/movement operation

I zoom in post as needed and don't worry about it too much, when using a full frame camera and 4k. Most of my family audience is going to watch it on an ipad or phone, not a large monitor, so if the zoomed footage looks good on my editing monitor, I go with it.

You might try a test at home, in not-so-good room light. Shoot some clips at 35-50 mm wide view of your daughter and the room, then zoom in in post to something closer to portrait framing for a portion of the clip, and see if the noise and resolution is acceptable.

When shooting the test clips, try setting the shot framing in camera with your daughter centered, then to the left and then right side of the frame and zoom in on each to see if there is much variation left to right. If the lighting is even, even a zoom to 30% of the original width may look ok.

Years ago, I tried to do this with small sensor camcorders. That is a no on zooming in post. In good light, one can do some zooming with a m43 sensor camera. In less than good light on a full frame camera in 4k, one can zoom a bunch in post. Note--we are talking family video here, not NetFlix, not commercial video.

The limitation is likely to be focus accuracy if your subject isn't the focus point, and variation in the scene lighting if the camera sets the exposure on a well lit part of the scene and your subject is in the shadows.

You can sort this out for yourself with a little experimenting at home.

Joe L

 

[winnegehetoch]

In Resolve Studio you have access to a host of extras, here an example:


Resolve Studio 20.0 example

For 4k material:

If a need minor scaling, it's done directly. Nobody notices.

If I need to do zoom and/or Ken Burns effects, I use a 2x Enhanced upscale, with following variable downscaling (variable "zoom"). It works better for my material. Nobody notices, if not done to excess (most important is detail acuity if zooming in to the background - most often need F11 or more).

Alternatives:

Lumix S5 with 12-bit 5.9k ProRES RAW material (I use Final Cut Pro 11), and in most cases no to minor upscaling (past pixel to pixel 4k) required. Using Atomos recorder. Nobody notices.

Fujifilm X-M5 12-bit 6.2k open gate 3:2 ProRES RAW. Using Atomos recorder. Nobody notices.

Walkabout camera Fujifilm X-M5 in open gate 6.2k (6240x4260) 10-bit 4:2:2 h265 HLG or FLog2 200 megabit/sec internal recording. Upscaling in post allows anything from oversampling to 4k to zoom into pixel-to-pixel 4k (a 15-45 mm (23-68 eqv) tiny kit zoom reaches 110 mm eqv in 4k cropped pixel to pixel (as an example). In effect a nice 23-110 mm zoom, without any upscaling in post. The h264 material is surprisingly versatile - ideal for travel walkabout. Nobody notices.

Regards

 

[Andrew S10]

You can crop in 2x without perceptible quality loss when supersampling UHD to 1080P.

 

[Gato Amarillo]

I'll agree -- this is simple enough to test for yourself.

I doubt you'll see much loss unless you go to an extreme zoom, and even less likely your audience will notice. But you know your standards and your audience.

 

[BobKo]

I know that you're asking about zooming and upscaling in Resolve, but the ability to pan around and zoom is exactly why I shoot most of my videos in 5.7K using my Panasonic GH6. It gives me some nice "elbow room" to shift the image around without losing 4K resolution. The GH6 (and the GH7 and probably the S5/S1) also have an "open gate" mode that shoots 5.8K at a 4:3 aspect ratio. That gives you quite a bit of freedom in post to move the image up/down left/right and still maintain 4K resolution. You can't go completely crazy with it, but having the flexibility is really nice.

 

[SillyPosition]

You can crop in 2x without perceptible quality loss when supersampling UHD to 1080P.

Thing is, I would like to preserve 4k. we do watch it on big screen tv, so I prefer 4K output

 

[SillyPosition]

I know that you're asking about zooming and upscaling in Resolve, but the ability to pan around and zoom is exactly why I shoot most of my videos in 5.7K using my Panasonic GH6. It gives me some nice "elbow room" to shift the image around without losing 4K resolution. The GH6 (and the GH7 and probably the S5/S1) also have an "open gate" mode that shoots 5.8K at a 4:3 aspect ratio. That gives you quite a bit of freedom in post to move the image up/down left/right and still maintain 4K resolution. You can't go completely crazy with it, but having the flexibility is really nice.

Yes, I guess that's the best option for such a case, If I had it. Thanks!

 

[SillyPosition]

In Resolve Studio you have access to a host of extras, here an example:


Resolve Studio 20.0 example

For 4k material:

If a need minor scaling, it's done directly. Nobody notices.

If I need to do zoom and/or Ken Burns effects, I use a 2x Enhanced upscale, with following variable downscaling (variable "zoom"). It works better for my material. Nobody notices, if not done to excess (most important is detail acuity if zooming in to the background - most often need F11 or more).

Alternatives:

Lumix S5 with 12-bit 5.9k ProRES RAW material (I use Final Cut Pro 11), and in most cases no to minor upscaling (past pixel to pixel 4k) required. Using Atomos recorder. Nobody notices.

Fujifilm X-M5 12-bit 6.2k open gate 3:2 ProRES RAW. Using Atomos recorder. Nobody notices.

Walkabout camera Fujifilm X-M5 in open gate 6.2k (6240x4260) 10-bit 4:2:2 h265 HLG or FLog2 200 megabit/sec internal recording. Upscaling in post allows anything from oversampling to 4k to zoom into pixel-to-pixel 4k (a 15-45 mm (23-68 eqv) tiny kit zoom reaches 110 mm eqv in 4k cropped pixel to pixel (as an example). In effect a nice 23-110 mm zoom, without any upscaling in post. The h264 material is surprisingly versatile - ideal for travel walkabout. Nobody notices.

Regards

Thanks.

For 4K material, 2x enhanced upscale seems to be a pro feature, altough I dont mind a single payment for pro if some good features that a novice like me can benefit from.

Does it increases amount of noise? Have you noticed any significant downsides to upscale & zoom?

I noticed that If I do anything like +30% zoom or so (transform is x1.3 or so) it starts to be very apparent that Im stretching it.

Also my video is taken at a dark show room (at least the one I tested it on, and recently filmed) so its even more noticeable.

 

[SillyPosition]

I understand that zooming in and then exporting to 4k does some sort of upscaling to the video, because the input was 4k to begin with, how severe is it to do so?

Its to the point where if its quite good, I think its much easier shooting wider being perfectly still, and do it all in post, getting a very smooth zoom/movement operation

I zoom in post as needed and don't worry about it too much, when using a full frame camera and 4k. Most of my family audience is going to watch it on an ipad or phone, not a large monitor, so if the zoomed footage looks good on my editing monitor, I go with it.

You might try a test at home, in not-so-good room light. Shoot some clips at 35-50 mm wide view of your daughter and the room, then zoom in in post to something closer to portrait framing for a portion of the clip, and see if the noise and resolution is acceptable.

When shooting the test clips, try setting the shot framing in camera with your daughter centered, then to the left and then right side of the frame and zoom in on each to see if there is much variation left to right. If the lighting is even, even a zoom to 30% of the original width may look ok.

Years ago, I tried to do this with small sensor camcorders. That is a no on zooming in post. In good light, one can do some zooming with a m43 sensor camera. In less than good light on a full frame camera in 4k, one can zoom a bunch in post. Note--we are talking family video here, not NetFlix, not commercial video.

The limitation is likely to be focus accuracy if your subject isn't the focus point, and variation in the scene lighting if the camera sets the exposure on a well lit part of the scene and your subject is in the shadows.

You can sort this out for yourself with a little experimenting at home.

Joe L

Thanks. Regarding audience you're right. But when I want to view it, I do care, same as we keep original raws or full sized 60mp jpegs or whatever not-so-needed-for-the-ordinary-guy

 

[winnegehetoch]

I assume render format is equal to source format - for instance 25fps 10-bit 4:2:2 HLG 2160p.

First:

ANY zoom-in (scaling above 100%) in any form and any method will decrease sharpness and enhance the visibility of existing noise to some degree.

Within reason: If the story is important to the viewers, nobody will notice. Tight editing will help a lot to "hide" imperfections. Long sequences with questionable quality can be detrimental, to general public joy - but imperfections usually have to be dramatic ;-)

Second:

ANY decent lense (note the term “decent”, which includes cheap) with the capability to zoom into required “crop” will be better than zooming/upscaling in post (for example using FF 100mm instead of FF 70mm followed by 1.4 times zoom in post). That will also simplify getting the best exposure for the actually selected framing, Which is harder to guarantee - depending on setting - with anything but minor zoom’s in post.

My motto has since the 1960’s been:

Get the image/shot/video whatever it takes. As long as it is in focus and sharp, most everything else doesn’t count, if the motif is actually "in frame". Since the 1980’s that has covered video too (whether film, tape color reel-to-reel, cartridge or other methods).

Indoor is seldom a problem, since low light mostly only covers what the audience is not supposed to see. ANY performance, that has to be visible by humans - even outdoor, deep night - has to have a certain minimum of light. Otherwise no one can see, what is going on. If a stage is involved, there’s “never” low light - but enough “available light” for everyone to enjoy the performance. Usually.

On my X-M5 (APS-C) my max ISO is set to 12.800 for video. On my S5 (FF) it is set to 25.600 ISO (images 51.200 ISO). ProRES RAW has special handling on especially the S5, but otherwise…. This guarantees a nice selection of decently “frozen action” also at night with standard travel zooms, where shot selection has to be flexible.

Noise is seldom a problem. What use is perfectly noise free, if everything in focus is a blur?

Biggest problem

Superscaling can be (has been in the cases I have used it) extremely time consuming. Sometimes, though, it’s the only way forward, if a specific take absolutely has to be used (talent dead, building demolished, once in a lifetime experience etc) ”requiring huge amounts of time “polishing” in post .

YMMV

Depending on your focus and exposure, your results may never warrant the time invested in superscaling compared to direct and simple transform zoom.

You pay "sloppy work" with extreme use of time for even short takes. It will seldom replace a somewhat correct framing at recording time.

Comment

Please note, that during Resolve 20 beta, there were some bruhaha about free upgrade. It was established, that version 20 was a free upgrade. Nothing was said about later major version upgrades (at the time of discussion).

I haven’t looked into the matter since.

I've used Resolve Studio since version 16, but my main solution is a combination of Final Cut Pro, Compressor and Motion - especially where ProRES RAW is involved. That will continue until Resolve supports ProRES RAW.

Regards

 

[winnegehetoch]

Render Speed examples

I was testing my new Atomos Ninja Phone for travel use today (no internal fan), so my computer had nothing to do. This will NEVER do, so it was immediately set to work on producing test results for various types of (super)scaling on Resolve Studio.

Single tests should be taken with a grain of salt, but the time involved - see below - is prohibitive, if ten, twenty or a hundred samples of various kinds had to be put through similar paces.

Resolve Studio 20.0.1

Final Cut Pro 11.1.1

MacOS Seqouoia 15.5

MacBook Pro 14 M1 Pro 16GB/1TB, external Thunderbolt 4 drives (Samsung 990 Pro 4TB). Cache/Backup work drive external Thunderbolt 4 (Samsung 980 Pro 2TB).

Source format:

2160p 25fps ProRES 422 HLG (01min 01sec 11 frames). 3.63 GByte

Target format:

2160p 25fps h265 10-bit 4:2:2 HLG 75 megabyte/sec

All tests on same machine.

Results:

Final Curt Pro 11.1.1 (example)

• NO scaling: 00min 31sec 597 MByte
• Direct 140% Transform zoom 00min 38sec 596 MByte

Render request on Resolve Studio 20.0.1 for this particular comparison:

• NO scaling: 00min 36sec 577 MByte
• Direct 1.4x Transform zoom: 00min 44sec 577 MByte

Final Cut Pro and Resolve Studio end up in the same ballpark. Apple a bit quicker, but not really significantly so in this case. None of the above renders make the fan noticed in any way.

• Superscale x2 default settings plus Fusion 0.7x Transform plus crop 2min 49s 576 MByte

GPU constantly pegged around 82-84%. No fan heard.

• Superscale x2 enhanced default settings plus Fusion 0.7x Transform plus crop 40min 49s 578 MByte

GPU constantly pegged around 97-99%. Fan working overtime.

As I said in a previous post, the “cost” in time using Superscale can be extreme in my view.

In simpler test cases, I perform a superscale x2 (enhanced), and place a simple overlapped zoom x2 version original version on the same timeline, subtracting the two video tracks. Rendering to 2160p output. The result clearly shows any “differences”. You decide, if the results are worth the investment of time.

Remarks:

“Use hardware accelleration if available” is checked

Procy, Cache etc. folders points to separate 2TB Thunderbolt 4 drive (Samsung 980 Pro) only used for this purpose (ample free space).

As always: YMMV

Regards