Bitrate calculations and implications for best quality

[qtonic]

The P4P uses the following bitrates to record at the various resolutions for H.264 and H.265.

H.265
C4K:4096×2160 24/25/30p @100Mbps
4K:3840×2160 24/25/30p @100Mbps
2.7K:2720×1530 24/25/30p @65Mbps
2.7K:2720×1530 48/50/60p @80Mbps
FHD:1920×1080 24/25/30p @50Mbps
FHD:1920×1080 48/50/60p @65Mbps
FHD:1920×1080 120p @100Mbps
HD:1280×720 24/25/30p @25Mbps
HD:1280×720 48/50/60p @35Mbps
HD:1280×720 120p @60Mbps

H.264
C4K:4096×2160 24/25/30/48/50/60p @100Mbps
4K:3840×2160 24/25/30/48/50/60p @100Mbps
2.7K:2720×1530 24/25/30p @80Mbps
2.7K:2720×1530 48/50/60p @100Mbps
FHD:1920×1080 24/25/30p @60Mbps
FHD:1920×1080 48/50/60 @80Mbps
FHD:1920×1080 120p @100Mbps
HD:1280×720 24/25/30p @30Mbps
HD:1280×720 48/50/60p @45Mbps
HD:1280×720 120p @80Mbps

Based on these I did some simple calculations to determine how much data rate should theoretically be available per pixel at each of the settings. To do this I used the total pixels per frame (based on resolution), then multiplied that by the frame rate to get the number of pixels per second. The bit rate was then divided by that number to calculate a ratio. The larger the ratio, the more of the bit rate should theoretically be available to each pixel displayed.

Before getting to the results I want to acknowledge that this comes without any technical understanding of how the codecs work or are optimized, for instance regarding frame rate efficiencies. Since the codecs work on multiple frames and changes between them, perhaps the frame rate has a less than linear relationship to quality vis-à-vis the bitrate. There are other factors I don't fully understand as well. This is only meant as a curiosity and to start a technical discussion. I appreciate knowing if anyone better understands how this might translate to real world results or how the data might affect your workflow decisions.

Results: (Note: the bitrate divided by total pixels per second is a very small number. I multiplied it by 10 million for more readability)

H.265

Ratio Resolution
11.303 720 @ 24
10.047 1080 @ 24
9.042 720 @ 30
8.038 1080 @ 30
7.912 720 @ 48
6.531 1080 @ 48
6.508 2.7K @ 24
6.330 720 @ 60
5.425 720 @ 120
5.224 1080 @ 60
5.206 2.7K @ 30
5.023 4K @ 24
4.710 C4K @ 24
4.019 4K @ 30
4.019 1080 @ 120
4.005 2.7K @ 48
3.768 C4K @ 30
3.204 2.7K @ 60

H.264

Ratio Resolution
13.563 720 @ 24
12.056 1080 @ 24
10.851 720 @ 30
10.173 720 @ 48
9.645 1080 @ 30
8.138 720 @ 60
8.038 1080 @ 48
8.010 2.7K @ 24
7.234 720 @ 120
6.430 1080 @ 60
6.408 2.7K @ 30
5.023 4K @ 24
5.006 2.7K @ 48
4.710 C4K @ 24
4.019 4K @ 30
4.019 1080 @ 120
4.005 2.7K @ 60
3.768 C4K @ 30
2.512 4K @ 48
2.355 C4K @ 48
2.009 4K @ 60
1.884 C4K @ 60

One of the results that stands out to me is 1080p @ 24. I already shoot 24 fps (H.265) most of the time, and my current workflow includes DaVinci Resolve lite which won't output 4K anyway. I had been shooting 4K or 2.7K thinking I'd have room to pan/zoom/tilt/etc, but if there is a real world quality difference based on the bitrate data above, I might convince myself to work straight 1080p from start to finish.

More testing and input are needed before making any conclusions. For instance, I'm curious if downscaling from 4K or 2.7K recovers any potential quality loss attributed to the bitrate, and I still wonder how much the codec is affected by frame rate.

Of course all of this is basically irrelevant if you definitely want a 4K final product, but hopefully it's interesting from a technical perspective at least.

 

[Tomas Wangen]

The P4P uses the following bitrates to record at the various resolutions for H.264 and H.265.
I already shoot 24 fps (H.265) most of the time, and my current workflow includes DaVinci Resolve lite which won't output 4K anyway.

Hi, interesting numbers.

I just have to mention that Davinci Resolve 12.5 (free) exports 4k fine, so this limitation has been removed. However, above 4k and hdr still requires a license.

As an example (i did post this before), this is h265 / 30 / 4k / dlog edited in Resolve 12.5

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

 

[qtonic]

Hi, interesting numbers.

I just have to mention that Davinci Resolve 12.5 (free) exports 4k fine, so this limitation has been removed. However, above 4k and hdr still requires a license.

Thanks for the correction. I'm glad to know that. Now hopefully they'll introduce H.265 support for free so I don't have to transcode outside of Resolve. I'd still generate optimized media anyway, but I'd rather do it all in Resolve.

 

[JimHarris]

Very interesting numbers. I would like to know if the numbers translate directly to picture quality. Someone with some time on their hands and a general curiosity needs to put them to the test.

 

[Roamer105]

One more thought regarding those numbers. If you shoot UHD and your target is 1080p, you have a much better compression ratio down-converting it from UHD (4 * 4.710 = 18.84). An additional advantage is the possibility to use interim UHD conversion to true 1080p 4-2-2 at 10bit since there is enough recorded information in UHD to do that. The resulting quality is much better than recording directly to 1080p.

 

[qtonic]

One more thought regarding those numbers. If you shoot UHD and your target is 1080p, you have a much better compression ratio down-converting it from UHD (4 * 4.710 = 18.84). An additional advantage is the possibility to use interim UHD conversion to true 1080p 4-2-2 at 10bit since there is enough recorded information in UHD to do that. The resulting quality is much better than recording directly to 1080p.

Honestly you lost me a little with this post, but it seems like just the sort of information I'm after. Could you elaborate a bit on the technical details if you have time? I appreciate it.

 

[Roamer105]

Honestly you lost me a little with this post, but it seems like just the sort of information I'm after. Could you elaborate a bit on the technical details if you have time? I appreciate it.

OK, here is why I would always record at the maximum resolution and bitrate regardless of where I want to end up:
If you record UHD (2160p) you encode (compress) exactly 4 times as much information as if you record 1080p. Both modes by encoder constraints record Chroma (color) information at 1/4 of the Luma (black/white) information. This is also referred to as chroma sub-sampling - generally an acceptable compromise since your eyes also have fewer color receptors than b/w receptors. Chroma sub-sampling of 4-2-0 means 4 bits of black/white information are clustered into 1 bit of color information. 4-2-0 also identifies exactly how the clustering occurs but that is more detail than required here. So in essence, UHD records as much color resolution as 1080p in 4-4-4 mode (each Luma bit also records color - there is no chroma sub-sampling in 4-4-4 mode). Of course the 1080p compression you identified has less loss than the UHD compression, but that is made up by the down-conversion process. Theoretically down-converting 2160p 4-2-0 (8-bit) should yield high quality 1080p 4-4-4 (10-bit). Of course, this all depends on the quality of the down-converter encoder. That is why I mentioned an interim format such as high quality cineform (RGB 12bit or YUV-10bit) or DNX-RGB 4-4-4).

 

[Roamer105]

Here is an example of processing from UHD to 1080p via DNX RGB 4-4-4 using Adobe Premiere Pro 2017.

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

Sent from my iPad using PhantomPilots

 

[qtonic]

OK, here is why I would always record at the maximum resolution and bitrate regardless of where I want to end up: If you record UHD (2160p) you encode (compress) exactly 4 times as much information as if you record 1080p...

Thanks so much for the technical information above. I will follow your advice, including your suggestion about the intermediate file format. So far I have been using DNxHR HQ which is 4:2:2 8 bit. My thinking was that with the source being 4:2:0 8 bit, that would be sufficient. I think I understand you correctly why in the case of conversion from 4K to 1080p I should be using something at least 4:4:4 10 bit to maximize retention of data. That helps me a lot in developing my workflow.

Nice video, by the way! I have some family in St. George. Great scenery out there.

 

[John Locke]

Here is an example of processing from UHD to 1080p via DNX RGB 4-4-4 using Adobe Premiere Pro 2017.

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

That's some beautiful scenery you captured. What kind of goggles are you using?

Also, can you comment on the range extender antenna's you installed on that controller. I assume that's a GL300C with a P4, correct? Those antenna paddles are inexpensive on Amazon, I'm just curious if you like them, and what kind of range you get in areas like in this video.

I noticed your pitch moves with the camera gimbal are somewhat abrupt. Have you played with the setting for "Pitch Smoothing". I've found that really helps mitigate the abruptness of the gimbal pitch move starts and stops. Sadly they omitted the "Pitch Smoothing" feature in Mavic, I'm hoping it's released in an update soon.

 

[Roamer105]

Here goes some info: I use a relatively cheap set of goggles that fit my iPhone 7. I believe they are "PasoNomi" Google Cardbox type goggles with 2-way adjustment capability. I modified them by drilling a hole through the phone cover so I can switch from FPV to Camera view.
I used an ND8 on this particular video. I don't like to go for too much motion blur as my real target resolution is UHD encoded HEVC Main-10 profile (10-bit color) at 50Mb/sec to be viewed on an LG 65" 4K OLED TV. Modern TV's all have excellent motion compensation, so the video even at 24fps is perfectly smooth. For Youtube, I down-converted the original as stated earlier. I have terrible up-load speed on my internet connection, so I don't bother uploading 4K. I use 1080p at 16Mb/s (default setting on PP/Media Encoder but with process to maximum depth enabled).
The antenna paddles (not amplified) are ALPHA 7dB at 2.4GHz antennae. They work very well. I've been flying out to 3 miles with no problem (100% signal) and they do penetrate reasonably well at closer distances.
As far as Pitch Smoothing - yes I know about that feature, but have not changed it (yet). I plan to play with that setting next time out to allow smoother head tracking.

Here is the flight log for this mission: Note the 100% signal score etc..
HealthyDrones.com - Innovative flight data analysis that matters

 

[Outtahere100]

Here goes some info: I use a relatively cheap set of goggles that fit my iPhone 7. I believe they are "PasoNomi" Google Cardbox type goggles with 2-way adjustment capability.

I'm curious about using the googles, do you have a background app to make the DJI App work properly? I looked at a set of Pasonomi goggles a friend has, and when I try to view the app on my phone, I can not see the center portion of the image. Great video BTW.

Thanks,

John

 

[Roamer105]

I am using Litchi FPV. I also use FPV Camera. In either case, I use my iPhone 7 inside the goggles.

My routine is using DJI GO to calibrate and set Camera Paramus and pre-flight checks then I kill that app and start either Litchi or FPV Camera put the phone into the goggles and start my flight. I calibrate my compass before every new flight.

Sent from my iPad using PhantomPilots

 

[skyeboysteve]

Regarding the two codecs, does H265 on the P4P encode 10-bit video? I thought I saw that assertion made in another forum, but can't find or confirm if this is really true.

 

[Roamer105]

Regarding the two codecs, does H265 on the P4P encode 10-bit video? I thought I saw that assertion made in another forum, but can't find or confirm if this is really true.

No, not according to the header information and by all means if it did, DJI would advertise that in BOLD LETTERS for sure!

 

[SRJ_Phantom_4_Pro]

Here goes some info: I use a relatively cheap set of goggles that fit my iPhone 7. I believe they are "PasoNomi" Google Cardbox type goggles with 2-way adjustment capability. I modified them by drilling a hole through the phone cover so I can switch from FPV to Camera view.
I used an ND8 on this particular video. I don't like to go for too much motion blur as my real target resolution is UHD encoded HEVC Main-10 profile (10-bit color) at 50Mb/sec to be viewed on an LG 65" 4K OLED TV. Modern TV's all have excellent motion compensation, so the video even at 24fps is perfectly smooth. For Youtube, I down-converted the original as stated earlier. I have terrible up-load speed on my internet connection, so I don't bother uploading 4K. I use 1080p at 16Mb/s (default setting on PP/Media Encoder but with process to maximum depth enabled).
The antenna paddles (not amplified) are ALPHA 7dB at 2.4GHz antennae. They work very well. I've been flying out to 3 miles with no problem (100% signal) and they do penetrate reasonably well at closer distances.
As far as Pitch Smoothing - yes I know about that feature, but have not changed it (yet). I plan to play with that setting next time out to allow smoother head tracking.

Here is the flight log for this mission: Note the 100% signal score etc..
HealthyDrones.com - Innovative flight data analysis that matters

Wow!! 846 feet!! Is that right? **** you were up there!

 

[Roamer105]

I have actually flown at the max hight (1630 feet or thereabouts) in some other more serious mountains- took me a while to get down from that altitude.

 

[marcinsow]

One more thought regarding those numbers. If you shoot UHD and your target is 1080p, you have a much better compression ratio down-converting it from UHD (4 * 4.710 = 18.84). An additional advantage is the possibility to use interim UHD conversion to true 1080p 4-2-2 at 10bit since there is enough recorded information in UHD to do that. The resulting quality is much better than recording directly to 1080p.

Hi,

That's interesting and it would mean I was also wrong thinking that 1080p would give me best quality (if we assume 1080 resolution was a target). Anyway, I have maybe dumb question, but I still may lack a bit in terms of processing workflow. Last time I did 4K by P4P and used that 4K clips in 1080p project (Final Cut Pro X), so they were all converted into 1080p upon export. Will this give me better bits-per-pixel ratio as well, or it is better to first convert clips to 1080p in external software (any suggestions?) and then go to edit project in FCPX?

Thanks!

Marcin

 

[marcinsow]

Hi, interesting numbers.

I just have to mention that Davinci Resolve 12.5 (free) exports 4k fine, so this limitation has been removed. However, above 4k and hdr still requires a license.

As an example (i did post this before), this is h265 / 30 / 4k / dlog edited in Resolve 12.5

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

Wow, I'm impressed how good it looks! I have also recorded some trees with lots of details in the branches (4K 30p with Phantom 4 Pro) but then once I processed in FCPX and uploaded to YouTube it got pixelized/artifacts in some moments. Someone told me that even with 100 Mbits in 4K it's still not enough for such fine details and then YouTube compression will make it worse. Even downscaling to 1080p did not help. But you uploaded to YouTube and it looks just great, no compression artifacts. Can it be just because you used h265? I did mine in h264. By the way, here is the link to the post with my video with issues if you're interested. Maybe your input will be helpful!

Best quality video settings in P4 Pro? 4K disappointing.

Thanks!

All the best!

Marcin

 

[Roamer105]

Hi,

That's interesting and it would mean I was also wrong thinking that 1080p would give me best quality (if we assume 1080 resolution was a target). Anyway, I have maybe dumb question, but I still may lack a bit in terms of processing workflow. Last time I did 4K by P4P and used that 4K clips in 1080p project (Final Cut Pro X), so they were all converted into 1080p upon export. Will this give me better bits-per-pixel ratio as well, or it is better to first convert clips to 1080p in external software (any suggestions?) and then go to edit project in FCPX?

Thanks!

Marcin

I would always do the down-convert as the very last step which means, edit in the highest and best quality, then export/down-convert. To enhance the ability to edit, I first do all the grading on the original footage in Davinci Resolve (14 - latest version) and export the graded "raw" footage to "MFX OP1A formatted "DNxHR 444 10 bit" or "DNxHR HQX 10 bit", both formats that edit well and are high quality. I use that material for actual editing in Premiere PRO and then final export using the Adobe Media Converter driven by PPro. I usually produce a high quality H265 (100mb - 10bit color and a Key Frame distance of 8) for my own consumption viewed on an LG 4K OLED TV. Using the interim format conversion and a reduced Key Frame distance to 8 almost completely eliminates the annoying "B-frame pulsing" produced by the lazy DJI compression engine. Note that in addition, I actually record using (Landscape) +1 sharpness and 0-0 for the rest in DCinelike at 24fps. The completed results are stunningly improved because the increased detail acts like a "kick-in-the-but" to the "lazy" DJI encoder and recording at 24fps allows it to squeeze all/most of that extra information into the compressed format. If it is too sharp in post production, reducing the sharpness a little in PPro works much better then trying to recover sharpness on washed out smudgy recordings.

Attached is a frame-grab from P3P video (UHD/24fps - Landscape mode DCinelike processed as described above).