OM-1 Video: h265 @ 25fps (77Mbit/sec) much more detail than h265 @ 50fps (152Mbit/sec)

[abgestumpft]

Hi,

I just received my OM-1 yesterday and was running some quick tests with video.
One thing was that I wondered about differences (e.g. in detail) between H265 and H264 and different frame rates ( they use different bitrates used: lower frame rates also use lower bitrates).

I was shooting a scene with lots of details from tripod (no camera movement).
To my surprise the h265 @ 25fps with 77Mbit/sec has much more detail than H265 @ 50fps with 152Mbit/sec and all H264 combinations I tested.

Here some screenshots directly from out of camera clips (no grading applied to HLG and Log footage).


h265 50fps HLG - 124530 kb/s


h265 50fps LOG - 131801 kb/s


h265 25fps HLG - 74196 kb/s


h265 25fps LOG - 74150 kb/s


h264 25fps - 98603 kb/s


h264 50fps - 197986 kb/s

I could understand why h265 shows more detail than h264 when using little lower bitrates. But I really wonder why h265 25fps 77 Mbit/sec shows much more detail than h265 50fps 152 Mbit/sec (twice the framerate and twice the bitrate). I run two more other tests (also with 60fps) and the results look similar to this.

Does anyone have an idea why h265 25fps looks better than h265 50fps?
Maybe someone with an OM-1 can test and confirm this?

 

[Interceptor121]

Image quality is mostly driven by the gop structure and the size and quality of the I frame

Do you know if both are the same codec and gop structure and subsampling? If you use mediainfo it will tell you and you can paste here and I will tell you what else you need to look for

 

[abgestumpft]

Image quality is mostly driven by the gop structure and the size and quality of the I frame

Do you know if both are the same codec and gop structure and subsampling? If you use mediainfo it will tell you and you can paste here and I will tell you what else you need to look for

With ffmpeg I get this (subsampling is 4:2:0 for all of them):

HLG:
Stream #0:0(jpn): Video: hevc (Main 10) (hvc1 / 0x31637668), yuv420p10le(pc, bt2020nc/bt2020/arib-std-b67), 3840x2160 [SAR 1:1 DAR 16:9], 124530 kb/s, 50 fps, 50 tbr, 180k tbn, 50 tbc (default)

Stream #0:0(jpn): Video: hevc (Main 10) (hvc1 / 0x31637668), yuv420p10le(pc, bt2020nc/bt2020/arib-std-b67), 3840x2160 [SAR 1:1 DAR 16:9], 74196 kb/s, 25 fps, 25 tbr, 90k tbn, 25 tbc (default)

OMLOG400:
Stream #0:0(jpn): Video: hevc (Main 10) (hvc1 / 0x31637668), yuv420p10le(pc, bt2020nc/bt2020/bt2020-10), 3840x2160 [SAR 1:1 DAR 16:9], 131801 kb/s, 50 fps, 50 tbr, 180k tbn, 50 tbc (default)

Stream #0:0(jpn): Video: hevc (Main 10) (hvc1 / 0x31637668), yuv420p10le(pc, bt2020nc/bt2020/bt2020-10), 3840x2160 [SAR 1:1 DAR 16:9], 74150 kb/s, 25 fps, 25 tbr, 90k tbn, 25 tbc (default)

 

[Interceptor121]

Image quality is mostly driven by the gop structure and the size and quality of the I frame

Do you know if both are the same codec and gop structure and subsampling? If you use mediainfo it will tell you and you can paste here and I will tell you what else you need to look for

With ffmpeg I get this (subsampling is 4:2:0 for all of them):

HLG:
Stream #0:0(jpn): Video: hevc (Main 10) (hvc1 / 0x31637668), yuv420p10le(pc, bt2020nc/bt2020/arib-std-b67), 3840x2160 [SAR 1:1 DAR 16:9], 124530 kb/s, 50 fps, 50 tbr, 180k tbn, 50 tbc (default)

Stream #0:0(jpn): Video: hevc (Main 10) (hvc1 / 0x31637668), yuv420p10le(pc, bt2020nc/bt2020/arib-std-b67), 3840x2160 [SAR 1:1 DAR 16:9], 74196 kb/s, 25 fps, 25 tbr, 90k tbn, 25 tbc (default)

OMLOG400:
Stream #0:0(jpn): Video: hevc (Main 10) (hvc1 / 0x31637668), yuv420p10le(pc, bt2020nc/bt2020/bt2020-10), 3840x2160 [SAR 1:1 DAR 16:9], 131801 kb/s, 50 fps, 50 tbr, 180k tbn, 50 tbc (default)

Stream #0:0(jpn): Video: hevc (Main 10) (hvc1 / 0x31637668), yuv420p10le(pc, bt2020nc/bt2020/bt2020-10), 3840x2160 [SAR 1:1 DAR 16:9], 74150 kb/s, 25 fps, 25 tbr, 90k tbn, 25 tbc (default)

The subsampling is identical so if there is a difference is in the gop structure

Normally higher frame rate footage looks sharper as long as you follow the 180 rule

You would need to run ffprobe to extract actual frame and measure however based on my experience I would guess that the gop structure is done in a way that the number of I frames per second does not actually change so the 50 fps has more prediction in it but yet it should not feel like it is dropping any quality

You do need to shoot a proper video chart though not uncontrolled outdoor footage

If you have a gdrive and put two clips each only 2 seconds long I can look into it for you for the gop structure

 

[abgestumpft]

Image quality is mostly driven by the gop structure and the size and quality of the I frame

Do you know if both are the same codec and gop structure and subsampling? If you use mediainfo it will tell you and you can paste here and I will tell you what else you need to look for

With ffmpeg I get this (subsampling is 4:2:0 for all of them):

HLG:
Stream #0:0(jpn): Video: hevc (Main 10) (hvc1 / 0x31637668), yuv420p10le(pc, bt2020nc/bt2020/arib-std-b67), 3840x2160 [SAR 1:1 DAR 16:9], 124530 kb/s, 50 fps, 50 tbr, 180k tbn, 50 tbc (default)

Stream #0:0(jpn): Video: hevc (Main 10) (hvc1 / 0x31637668), yuv420p10le(pc, bt2020nc/bt2020/arib-std-b67), 3840x2160 [SAR 1:1 DAR 16:9], 74196 kb/s, 25 fps, 25 tbr, 90k tbn, 25 tbc (default)

OMLOG400:
Stream #0:0(jpn): Video: hevc (Main 10) (hvc1 / 0x31637668), yuv420p10le(pc, bt2020nc/bt2020/bt2020-10), 3840x2160 [SAR 1:1 DAR 16:9], 131801 kb/s, 50 fps, 50 tbr, 180k tbn, 50 tbc (default)

Stream #0:0(jpn): Video: hevc (Main 10) (hvc1 / 0x31637668), yuv420p10le(pc, bt2020nc/bt2020/bt2020-10), 3840x2160 [SAR 1:1 DAR 16:9], 74150 kb/s, 25 fps, 25 tbr, 90k tbn, 25 tbc (default)

The subsampling is identical so if there is a difference is in the gop structure

Normally higher frame rate footage looks sharper as long as you follow the 180 rule

I guess in my case (almost static scene from tripod) this does not make any difference.
For all the clips I used the same exposure settings, I just changed the codec/framerate/picture mode.

You would need to run ffprobe to extract actual frame and measure however based on my experience I would guess that the gop structure is done in a way that the number of I frames per second does not actually change so the 50 fps has more prediction in it but yet it should not feel like it is dropping any quality

Not sure if I did this right, when counting "pict_type=I" and "key_frame=1" in ffprobe output:

For 25fps Clip:
Duration: 00:00:10.56, start: 0.000000, bitrate: 78686 kb/s
Stream #0:0(jpn): Video: hevc (Main 10) (hvc1 / 0x31637668), yuv420p10le(pc, bt2020nc/bt2020/arib-std-b67), 3840x2160 [SAR 1:1 DAR 16:9], 74196 kb/s, 25 fps, 25 tbr, 90k tbn, 25 tbc (default)

ffprobe.exe -select_streams v:0 -show_frames "P3160318.MOV" | find /i "key_frame" /c
264

ffprobe.exe" -select_streams v:0 -show_frames "P3160318.MOV" | find /i "key_frame=1" /c
22

ffprobe.exe -select_streams v:0 -show_frames "P3160318.MOV" | find /i "pict_type=I" /c
22



For 50fps Clip:
Duration: 00:00:10.56, start: 0.000000, bitrate: 131148 kb/s
Stream #0:0(jpn): Video: hevc (Main 10) (hvc1 / 0x31637668), yuv420p10le(pc, bt2020nc/bt2020/arib-std-b67), 3840x2160 [SAR 1:1 DAR 16:9], 124530 kb/s, 50 fps, 50 tbr, 180k tbn, 50 tbc (default)

ffprobe.exe -select_streams v:0 -show_frames "P3160316.MOV" | find /i "key_frame" /c
528

ffprobe.exe -select_streams v:0 -show_frames "P3160316.MOV" | find /i "key_frame=1" /c
22

ffprobe.exe -select_streams v:0 -show_frames "P3160316.MOV" | find /i "pict_type=I" /c
22

Calculating Keyframes:

264 / 22 = 12 -> every 12 frames a keyframe for 25fps clip
528 / 22 = 24 -> every 24 frames a keyframe for 50fps clip
For another 60fps clip I tested we have every 30 frames a keyframe.

So looks like the OM-1 is writing every 1/2sec a keyframe, no matter what framerate we're running. The same also applies for the H264 clips (25fps and 50fps). So either the number of keyframes has high impact on h265 quality, or it's not related to the keyframe at all...

You do need to shoot a proper video chart though not uncontrolled outdoor footage

I don't have a video chart (surprise )
But I think the difference is obvious. I don't want to measure the resolution

If you have a gdrive and put two clips each only 2 seconds long I can look into it for you for the gop structure

If what I did was wrong above I will see if I find to create some short clips tomorrow, thanks!

 

[Interceptor121]

So this confirmed what I say however you need the size of the I frane so ask ff probe to output to file and check I frame size

i think they will be identical

 

[abgestumpft]

So this confirmed what I say however you need the size of the I frane so ask ff probe to output to file and check I frame size

i think they will be identical

I have no what you mean in detail
And: what would be the finding in case they are the same or different size?

 

[Interceptor121]

So this confirmed what I say however you need the size of the I frane so ask ff probe to output to file and check I frame size

i think they will be identical

I have no what you mean in detail
And: what would be the finding in case they are the same or different size?

Frame size determines quality