A crazy thought: Shoot ProRes HQ, but covert to h265 in post?!

[PJD]

I should have titled this post "Another crazy thought ..." since it's not my first, but whatever.

For no-budget video shoots (for example, the MOPAR dragstrip races my friend & I recently shot), where we're not being paid -- so there's no budget for buying additional big storage disks for archiving and backup -- I'm thinking I might start shooting them in 6K "ProRes HQ" mode (the S5iiX's highest-quality video recording mode), and then later use Davinci Resolve Studio (the $300 US version I own) to convert the ProRes HQ footage to 6K "Main 4:2:2 10" (10-bit) h265 HEVC @ 250,000 kilobits/sec. This can be done using either Davinci's DELIVER page or via its "Media Management" feature in the File menu.

The resulting h265 files look & sound fantastic and consume about 1/5th the disk space as the original ProRes HQ files.

Caution: Be sure to set the Davinci's DELIVER page "Browse" button (or Media Management) to render out to a different directory/folder than the one where your original ProRes files are stored, so you don't inadvertently delete/overwrite the original files. Later (perhaps after editing) you can decide if you no longer need the original files and delete them, but not just yet.

To render highest-quality h265 files in Davinci, before importing the source files, open Davinci and set the project Master Settings to the exact same resolution & frame rate as the original ProRes HQ files. Don't color-correct or apply a LUT in Davinci or modify the imported ProRes HQ clips in any way prior to rendering to h265 (so changes aren't baked-into the rendered h265 clips). In the DELIVER page, Restrict the "Quality" setting to a value higher than the camera's highest h265 bitrate (such as 250,000 Kb/s), turn off "Optimize for speed", turn on "Multi-pass encode", turn on "Render at source resolution", set "Render" to "Individual clips", set "Format" to "Quicktime H.265", enable "Use hardware acceleration if available", and set "Encoding profile" to "Main 4:2:2 10". In the DELIVER page (or Media Management) Audio tab, set to the Codec to "Linear PCM", and Bit Depth to "24". In the DELIVER page File tab, set "Filename uses" to "Source name" (so the h265 files will have the same name as the original ProRes files, unless you have a good reason to do otherwise).

On my old MacBook Pro M1 laptop, Davinci performs the conversions (renders) slightly slower than real time. IOW, 1 hour of footage takes about 1.25 hours to render. A faster computer will do it faster. For example, the 1.5 hrs of "ProRes 422" 4K DCI 60p footage I shot at the MOPAR races took about 2 hours to convert on my relatively slow M1 MBP13. The original ProRes 422 files require about 685 gigabytes of storage, whereas the files converted to high-quality h265 total about 133 gigabytes. For me, this makes a very useful difference in the time & cost of making (and maintaining) backups of the files to 4 sets of hard disk drives (3 stored locally, 1 stored remotely). Also, the now much smaller files easily fit on my SSD editing drive, and play smooth as silk in Davinci for editing, color-correcting, audio sweetening, etc.

Also, instead to rendering out to high-quality h265, some folks might prefer to render out to a different high-quality codec, in which case go for it.

Having said all that, I'm assuming ProRes HQ video rendered out from Davinci as "Main 4:2:2 10" h265 files @ 250,000 kilobits/sec will be at least as good quality (if not better) than the camera's best-quality h265 (4:2:0, 10-bit, 200,000 kilobits/sec). However, I could be wrong. If anyone has an opinion about any of this, please reply and/or smack some sense into me. Thanks!

 

[Nevyn72]

I should have titled this post "Another crazy thought ..." since it's not my first, but whatever.

For no-budget video shoots (for example, the MOPAR dragstrip races my friend & I recently shot), where we're not being paid -- so there's no budget for buying additional big storage disks for archiving and backup -- I'm thinking I might start shooting them in 6K "ProRes HQ" mode (the S5iiX's highest-quality video recording mode), and then later use Davinci Resolve Studio (the $300 US version I own) to convert the ProRes HQ footage to 6K "Main 4:2:2 10" (10-bit) h265 HEVC @ 250,000 kilobits/sec. This can be done using either Davinci's DELIVER page or via its "Media Management" feature in the File menu.

The resulting h265 files look & sound fantastic and consume about 1/5th the disk space as the original ProRes HQ files.

Caution: Be sure to set the Davinci's DELIVER page "Browse" button (or Media Management) to render out to a different directory/folder than the one where your original ProRes files are stored, so you don't inadvertently delete/overwrite the original files. Later (perhaps after editing) you can decide if you no longer need the original files and delete them, but not just yet.

To render highest-quality h265 files in Davinci, before importing the source files, open Davinci and set the project Master Settings to the exact same resolution & frame rate as the original ProRes HQ files. Don't color-correct or apply a LUT in Davinci or modify the imported ProRes HQ clips in any way prior to rendering to h265 (so changes aren't baked-into the rendered h265 clips). In the DELIVER page, Restrict the "Quality" setting to a value higher than the camera's highest h265 bitrate (such as 250,000 Kb/s), turn off "Optimize for speed", turn on "Multi-pass encode", turn on "Render at source resolution", set "Render" to "Individual clips", set "Format" to "Quicktime H.265", enable "Use hardware acceleration if available", and set "Encoding profile" to "Main 4:2:2 10". In the DELIVER page (or Media Management) Audio tab, set to the Codec to "Linear PCM", and Bit Depth to "24". In the DELIVER page File tab, set "Filename uses" to "Source name" (so the h265 files will have the same name as the original ProRes files, unless you have a good reason to do otherwise).

On my old MacBook Pro M1 laptop, Davinci performs the conversions (renders) slightly slower than real time. IOW, 1 hour of footage takes about 1.25 hours to render. A faster computer will do it faster. For example, the 1.5 hrs of "ProRes 422" 4K DCI 60p footage I shot at the MOPAR races took about 2 hours to convert on my relatively slow M1 MBP13. The original ProRes 422 files require about 685 gigabytes of storage, whereas the files converted to high-quality h265 total about 133 gigabytes. For me, this makes a very useful difference in the time & cost of making (and maintaining) backups of the files to 4 sets of hard disk drives (3 stored locally, 1 stored remotely). Also, the now much smaller files easily fit on my SSD editing drive, and play smooth as silk in Davinci for editing, color-correcting, audio sweetening, etc.

Also, instead to rendering out to high-quality h265, some folks might prefer to render out to a different high-quality codec, in which case go for it.

Having said all that, I'm assuming ProRes HQ video rendered out from Davinci as "Main 4:2:2 10" h265 files @ 250,000 kilobits/sec will be at least as good quality (if not better) than the camera's best-quality h265 (4:2:0, 10-bit, 200,000 kilobits/sec). However, I could be wrong. If anyone has an opinion about any of this, please reply and/or smack some sense into me. Thanks!

It is a recompression, bit I expect you definitely get better quality like this.

4:2:2 has twice as much color detail as 4:2:0

I haven't done test's with it yet but I expect excellent quality.

And if you would use it for greenscreen than it is a no-brainer

 

[PJD]

A few additional details (with video, there are always more details ...)

- I haven’t timed it, but Davinci’s Media Management feature might be slower than its DELIVER page? Maybe because on the DELIVER page you see the frames in the Preview window while they’re being rendered it just seems faster. Like I said, I haven’t timed the two methods to compare, but just an FYI.

- On they DELIVER page, also enable “Retain sub-black & super-white data”. You want all the image data from the original source video to be available for re-compression.

- If stereo audio was recorded in-camera, and you only want 1 track of stereo audio to be rendered out to the h265 file (instead of 4 mono tracks, 2 of which are empty): On the EDIT page, in the timeline change the audio type of audio track 1 to stereo (right -click on the far left of the audio track 1 header). Turn off Linked Selection (chain button at top of timeline). Select all the clips’ channel 2-4 audio (but not channel 1) and delete them (press the delete key). Then turn Linked Selection on again using the chain button. Then select all the clips (command-A) and right click one of them and choose Clip Attributes. In the Audio tab, set Format to Stereo, and Source Channel to “Embedded Channel 1” and “Embedded Channel 2”. Click OK. When you later use the DELIVER page as described previously, the resulting h265 video clips will have a single stereo audio track. That was easy, right? ;-)

 

[CharlesH]

A couple of thoughts. The best converter I've found for this kind of thing is Compressor. I know DaVinci will work, but if it were me I'd use Compressor.

By going from ProRes to h.265 you are going from an All-I format to a LongGOP format. ProRes is for editing, LongGOP is for delivery. If it was me I'd stick with ProRes until final delivery. I would be concerned with high motion scenes where All-I captures every frame and LongGOP does its best to recreate the action. As a test, shoot a picket fence from a moving car with both formats; All-I will look better. Here is a video that goes over this.



Also, you might want to delivery in h.264 if there is quite a bit of motion.

 

[Nevyn72]

A couple of thoughts. The best converter I've found for this kind of thing is Compressor. I know DaVinci will work, but if it were me I'd use Compressor.

By going from ProRes to h.265 you are going from an All-I format to a LongGOP format. ProRes is for editing, LongGOP is for delivery. If it was me I'd stick with ProRes until final delivery. I would be concerned with high motion scenes where All-I captures every frame and LongGOP does its best to recreate the action. As a test, shoot a picket fence from a moving car with both formats; All-I will look better. Here is a video that goes over this.



Also, you might want to delivery in h.264 if there is quite a bit of motion.

I Think @PJD wants to keep the 'originals' for when changes are in order, than it might be useful.

For me: to much hassle.
I shoot ProRes only when I want 6K 4:2:2 for greenscreen or 'lots of zooming', after finishing the project I never keep the originals, with the exception of a few shots I might need for another project in the future.

Also: If I don't need 6K and want very good quality due to movement, I'll use 4K All-I.

In our latest vacation, which unfortunately ends today, we shot a lot of footage for a few vlogs.

Last year I shot all in 4K all-Intra, this year I shoot this all in 4K 4:2:2 L-gop. Just as good for this purpose, and I don't need the external SSD connected, makes it easier to shoot 'lower profile'.

Why would H264 be better than H265?

 

[PJD]

A couple of thoughts. The best converter I've found for this kind of thing is Compressor. I know DaVinci will work, but if it were me I'd use Compressor.

By going from ProRes to h.265 you are going from an All-I format to a LongGOP format. ProRes is for editing, LongGOP is for delivery. If it was me I'd stick with ProRes until final delivery. I would be concerned with high motion scenes where All-I captures every frame and LongGOP does its best to recreate the action. As a test, shoot a picket fence from a moving car with both formats; All-I will look better. Here is a video that goes over this.



Also, you might want to delivery in h.264 if there is quite a bit of motion.

Thanks for your comments Charles.

Compressor 4.8 doesn't appear to be able to output h265 higher than 4K UHD (3840x2160) resolution, unless I'm missing something?

In my first post above I'm proposing (especially for non-paying, non-critical productions) shooting S5iiX 6K "ProRes HQ" (which is 4:2:2 10-bit) and then converting it to very high-quality 6K "Main 4:2:2 10" h265 HEVC using Davinci Resolve Studio. I plan to use the h265 files for editing & archiving, and throw away the original, gigantic ProRes HQ files. Depending on what data rate is chosen for the h265 renders, the h265 files will be typically be much smaller than the original ProRes HQ files, and _may_ be higher-quality than h265 and h264 the camera can record itself. The latter part is where I may be mistaken.

Tomorrow I'll shoot some tests using the camera locked on a tripod aimed at a "target" subject rotating on a record turntable, so I can record exactly-repeatable motion. I'll record h264, h265, ProRes 422, and ProRes HQ, and then compare them. Stay tuned.

 

[CharlesH]

Tomorrow I'll shoot some tests using the camera locked on a tripod aimed at a "target" subject rotating on a record turntable, so I can record exactly-repeatable motion. I'll record h264, h265, ProRes 422, and ProRes HQ, and then compare them. Stay tuned.

Good, that addresses my concern. The other thing I don't know, if you shoot in ProRes and then convert it to h.265 is the good motion preserved? Speculate that ProRes works better than h.265 in your motion test, what happens if you convert that ProRes to h.265 with Resolve? Does it look more like the original ProRes, or more like the original h.265?

By the way, I always film with an All-I format and then deliver in h.265. But I don't film drag races, so I don't have that much high speed motion to deal with.

 

[PJD]

Attached are the first of two sets of frames grabbed from video. These are extremely boring images, so be sure to look at other folks' photos of pretty flowers, cool trucks, landscapes, etc. elsewhere on this forum to cure any brain damage possibly caused by viewing these test images.

The filenames of each image describe the S5iiX video mode used to record the original video clip. This first set is a "picket fence" simulation. A white card with vertical black stripes was attached to a phonograph turntable spinning at 33.3RPM (the turntable's slowest speed). If nothing else, the camera's rolling shutter is amply demonstrated by the way the moving card is tilted. But we all know the S5iiX has a rolling shutter, so no news there. In fact, there's no news here at all really.

Each video clip was imported into an empty Davinci Resolve Studio project with the same resolution & frame rate as the video clip. The only color correction done was applying the VLOG LUT, and doing a white balance using the eyedropper tool to click on the middle of the white card. Frames were exported out of Davinci as 16-bit TIFFs. They were converted to JPEG using Pixelmator Pro and saved at 80% quality.

Please excuse a couple of "typos" written on my paper labels. For example, ProRes clips are not "MOV", and HEVC 6K is 17:9 not 16:9. I was tired.

Later, as Charles suggests, I convert the ProRes HQ clips to high-quality AVC (h264) and HEVC (h265) and compare them to the original h264 & h265 video clips.

I'll post the 2nd set of frame grabs in a separate post. The 2nd set features a colorful "target" with a variety of detail, and was shot looking down at the turntable so there's more than one rotation speed (diameter) visible. Stay tuned for that.







===

 

[PJD]

Just to add: The lens was in manual focus mode, focussed on the white card when it wasn't moving. The DOF changes between images resulted from adjusting the iris to keep exposure consistent. In hindsight I could have adjusted ISO instead, but that would changed noise levels. A better-designed test would keep the iris & ISO constant, and vary the light level instead. My apologies.

 

[Nevyn72]

Nice test, but for me it is hard to see which one are the best. What do you think?

Checking the writing, the last ones look 'thicker', but these didn't have movement and might just be because you did more pressure on the pencil...

 

[Nevyn72]

Nice test, but for me it is hard to see which one are the best. What do you think?

Checking the writing, the last ones look 'thicker', but these didn't have movement and might just be because you did more pressure on the pencil...

Stupid me, thats because of the different DOF

 

[PJD]

Here's my 2nd set of test images, this time featuring a more colorful target, and varying amounts of motion blur depending on where a detail is in relation to the center of the spinning turntable, frame rate, etc. Again, the filenames describe the original S5iiX video recording format for each image. I don't have any conclusions yet. As noted above, this isn't a perfect test, and doesn't tell us much we don't already know. Also as noted above, next I'll convert the ProRes HQ video clips to high-quality AVC (h264) and HEVC (h265) video and compare them to the original h264 & h265 video clips. I don't know yet if results will be useful. Cheers.






===

 

[CharlesH]

These are extremely boring images, so be sure to look at other folks' photos of pretty flowers, cool trucks, landscapes, etc. elsewhere on this forum to cure any brain damage possibly caused by viewing these test images

Good images, I'm a video guy.

But I am having trouble making sense of the images. The picket fence effect from LongGOP is a video effect, and I've not considered how it would show up with frame grabs. What might you expect? If you just play the video can you see any LongGOP effect? Or play the ProRes side by side with the HEVC, do they look the same?

 

[PJD]

Good images, I'm a video guy.

But I am having trouble making sense of the images. The picket fence effect from LongGOP is a video effect, and I've not considered how it would show up with frame grabs. What might you expect? If you just play the video can you see any LongGOP effect? Or play the ProRes side by side with the HEVC, do they look the same?

I realized after I posted my "picket fence" simulation images they might not have much value as frame grabs. I'll put together a side-by-side video edit comparing clips with matching frame rates. It'll be interesting to see if I post the edit on Vimeo what their compression does to it.

Meanwhile, I like how the S5iiX "ProRes HQ" 4K 60p looks, especially for scenes with relatively fast motion. I bet Compressor will be able to convert it to really good-looking HEVC h265 4K 60p or 30p. TBD.

Another meanwhile: This event is happening in my neighborhood today. I hope to take some photos & video of it:
https://www.multnomahvillage.org/event-details/multnomah-days-festival-parade

 

[PJD]

I realized after I posted my "picket fence" simulation images they might not have much value as frame grabs. I'll put together a side-by-side video edit comparing clips with matching frame rates. It'll be interesting to see if I post the edit on Vimeo what their compression does to it. ...

I'm going to re-shoot these two clips again. I want to double-check that the lens is focussed on the white card, and also to see if the results are consistently repeatable. In this split-screen, the 4K 4096x2160 60p AVC clip is on top; the 4k 4096x2160 60p ProRes HQ clip is on the bottom.

Is this split-screen presentation "format" good? When I reshoot the clips, I'll record more revolutions of the turntable so during playback there's more to watch without having to hit repeat. As usual, click the gear icon to watch in 4K, etc.

As noted on the Vimeo page, this is "A 'picket fence' simulation. For evaluating codec motion characteristics ONLY, not color, noise, etc. Both clips shot with a 180-degree shutter, and edited in Davinci Resolve in a 4K DCI 60p timeline, with saturation set to zero (0). ...'"

Again, I haven't made any conclusions yet, other than thinking I need to reshoot these clips to double-check my work.

 

[PJD]

Here are the re-shot 4K 60p AVC & ProRes HQ clips in a new timeline. The lens is definitely focussed on the stripes on the white card. Note in the edit I swap back & forth between which clip is shown top/bottom.

One thing that really stands out with all the 60p clips (AVC & ProRes) are how the vertical stripes on the card are oddly "doubled". The actual card doesn't look like that (see the screen grab image below). The clips I shot in 30p don't have that effect; i.e. the stripes aren't doubled. I'm not sure what causes that. Any thoughts?

 

[CharlesH]

One thing that really stands out with all the 60p clips (AVC & ProRes) are how the vertical stripes on the card are oddly "doubled". The actual card doesn't look like that (see the screen grab image below). The clips I shot in 30p don't have that effect; i.e. the stripes aren't doubled. I'm not sure what causes that. Any thoughts?

Strange. Just a thought, what is the light source? Is the motion blur maybe seeing the light cycle? I might try upping the shutter speed enough to get rid of the motion blur. Or make sure you have a flicker free light source.

(It is not a sin to shoot something other than a 180 degree shutter.)

 

[PJD]

Strange. Just a thought, what is the light source? Is the motion blur maybe seeing the light cycle? I might try upping the shutter speed enough to get rid of the motion blur. Or make sure you have a flicker free light source. (It is not a sin to shoot something other than a 180 degree shutter.)

Ah, I bet you're right: It's probably the lights. In each of the 6 times I recorded the 60p clips (3 AVC & 3 ProRes) I used a different light source. LED, then KinoFlo fluorescent, and lastly incandescent. Tomorrow I'll record 60p clips using natural daylight, to see if the doubled vertical lines go away. Thanks!

 

[PJD]

Thanks to CharlesH I think I finally have a "picket fence" simulation which might be useful. Please ignore my previous attempts. This test is illuminated by natural daylight, and as a result the vertical black stripes are no longer incorrectly "doubled". Instead they look nicely blurred, as expected. The doubling was caused by the interaction between the 180 degree shutter angle used with 60p (1/120th sec) vs. the 60-cycle AC powering the light fixtures used in the previous versions of the test. Also attached below is a frame grab showing what the paper "fence" looks like when NOT revolving. My initial conclusion is that both (AVC 4K 60P APS-C 422-10-I-H) and (ProRes-HQ 4K 60P APS-C 422-10-I) handle this type of motion equally well.

 

[CharlesH]

My initial conclusion is that both (AVC 4K 60P APS-C 422-10-I-H) and (ProRes-HQ 4K 60P APS-C 422-10-I) handle this type of motion equally well.

Good work. Nicely done.