I have a couple of questions regarding intra-frame codecs for editing purposes (in Premiere Pro).
UPDATE (01/24/20). More details of the original issue after this:
I seem to have encountered a dead end. While transcoding I noticed that the source file was tv, bt2020nc/bt2020/smpte2084 (I mean range, space, etc.), and when trying to transcode with the command specified down below (after this UPDATE) FFmpeg didn't say anything about those settings. I started to do some test giving FFmpeg input and output commands (-color_space 1, -color_trc 16, -colorspace 9, etc), and now it started to show the correct values while transcoding (bt2020nc...).
The command line for that test was:
ffmpeg -color_range 1 -color_primaries 9 -colorspace 9 -color_trc 16 -channel_layout 63 -i input.mkv -map 0:0 -c:v dnxhd -profile:v dnxhr_444 -pix_fmt yuv444p10le -color_range 1 -color_primaries 9 -colorspace 9 -color_trc 16 -acodec pcm_s24le -ar 48000 -ac 6 -channel_layout 63 -map 0:2 -hide_banner output.mxf
I read that those commands only add the flags to the container, but doesn't transcode or do anything to the stream itself. Like I said, while transcoding with that command line it said yuv444p10le (tv /bt2020nc,bt2020,smpte2084), the same as the source file, but when running FFprobe it still says yuv444p10le(bt709/unknown/unknown, progressive). The rest is fine (correct DNxHR profile, fps, resolution).
My guess was that those commands were only adding flags that maybe FFprobe and players don't actually read or use (the videos still look very different from the original footage), so I added a command using colorspace filter for actual conversion:
ffmpeg -color_range 1 -color_primaries 9 -colorspace 9 -color_trc 16 -channel_layout 63 -i input.mkv -map 0:0 -c:v dnxhd -vf format=yuv420p10le,colorspace=all=bt2020:trc=smpte2084:range=tv:primaries=bt2020:space=bt2020ncl -profile:v dnxhr_444 -pix_fmt yuv444p10le -color_range 1 -color_primaries 9 -colorspace 9 -color_trc 16 -acodec pcm_s24le -ar 48000 -ac 6 -channel_layout 63 -map 0:2 -hide_banner output.mxf
With that, I couldn't make past the initial checks because it says "unsupported input transfer characteristics 16 (smpte2084). No matter what I do with the colorspace filter, the output is always the same. The same happens if I delete the "trc=smpte2084" line. Anyway, that's the only error message I'm getting now, but seems like a dead end if it's not supported. And if I don't use the colorspace filter and only the flags (-color_trc...), it transcodes without issues, but FFprobe says it's BT709 (and behaves like a BT709 video while comparing it to the original footage using VLC).
QUESTIONS: Is SMPTE2084 really unsupported? Any workaround for this? Is it really possible to have BT2020 and true HDR with DNxHR 444 10 bits in a MXF container using FFmpeg? If not, any suggestions? Already tried with Adobe Media Encoder 2020, same results, but you don't have much settings to tweak there.
Thanks in advance.
A LITTLE CONTEXT:
I ripped a HDR movie to a MP4 container some months ago, using H.264 as codec (high 10, level 5, UHD, YUV, subsampling 4:2:0). The video looks great, as it should. Now I want to edit a trailer for that movie (I edit in Adobe Premiere Pro, current version), and for fast playback I need to work with proxies or transcode the source file to an intra-frame non long GOP file and use it as the source file (hopefully as lossless as that transcoding step can be). I tried for several days to transcode the source file to DNxHR 444 10 bits (using FFmpeg and then Adobe Media Encoder), but the result was always a video with the colors messed up (sometimes very washed out, sometimes over saturated).
FFprobe of the resulting DNxHR file said that the color space was BT709 (source file is obviously BT2020), and I don't know why. The transcoding involved upsampling since the source file is 4:2:0 and DNxHR doesn't support it, but I tried upsampling to 4:4:4 and also to 4:2:2, and both of those files looked exactly the same to me, and very different from the original footage (so, I don't think upsampling is the cause of the color issue, but maybe the apparent color space change or something wrong with the metadata). The results were the same when transcoding with Adobe Media Encoder. Anyway, I seem to have given up transcoding to DNxHR and use it as the source file, unless someone has an idea of what's causing this problem. I could have worked with the source file for exporting and DNxHR LB for proxies, but there were sync issues (between source file and proxy) that defeated all purposes while editing. Prores is out of the picture, sync issues were worse (several seconds of delay).
For the record, the command used that didn't work as expected (color wise) is:
ffmpeg -channel_layout 63 -i input.mkv -map 0:0 -c:v dnxhd -vf "scale=in_range=limited:out_range=full" -color_range 2 -profile:v dnxhr_444 -pix_fmt yuv444p10le -acodec pcm_s24le -ar 48000 -ac 6 -channel_layout 63 -map 0:2 -hide_banner output.mxf
I also tried without the commands "scale=in_range=limited:out_range=full" and "-color_range 2", with same results. Always used FFmpeg latest version, and I'm working in Windows 10 Pro, latest drivers and latest Klite codec pack. Video files were compared with Mediainfo, FFprobe, and visually with VLC.
Well, like I said, I'm giving up using DNxHR as the source file for my project (it would have been ideal since it doesn't have sync issues with the DNxHR proxies, and file size is not a problem for me). A user here at the forums suggested transcoding and use H.264 intra-frame as source file, which I didn't know was an option (I didn't know H.264 was capable of intra-frame, my bad). I'm aware that one should avoid unnecesary transcoding steps, but I can't work with a H.264 UHD HDR source file (ultra slow playback), and the sync issues with proxies, no matter the codec, make it impossible to make accurate cuts. So, bottom line, I need to find a way to fix the color issue when transcoding to DNxHR, or try with an inter-frame codec that's not DNxHR and that's capable of preserving all the HDR info (and then see if it doesn't have sync issues. I'm assuming that those may dissapear when using intra-frame for both source file and proxies).
NOTE: When importing the DNxHR 444 file to Premiere Pro and looking at the Lumetri scopes tab, you can tell that the colors are clipped at 100 nits, like a regular SDR video. So apparently the color space was really reduced to BT709, and I don't know why. The H.264 source file behaves as expected, with colors going past 100 nits.
1) Is H.264 intra-frame a good format for editing with a good playback performance?
2) If H.264 intra-frame is a good option, what would be the advantages of the DNxHR codec (or Prores) over it for editing purposes? Everybody suggests DNxHD/DNxHR or Prores as intermediate codecs, but if intra-frame H.264 has the same advantages for editing and supports HDR, what would be the reason to choose another codec like DNxHR?
3) Any ideas on what could be the cause of the colors not transcoding correctly from the H.264 source file to the DNxHR 444 10 bits one? The command looks ok to me, but FFprobe output says the DNxHR video is BT709, while with the source file it says BT2020. Like I said, apparently there's something wrong with the transcoding process regarding metadata or color space.
4) I haven't tried to transcode the source file to a DNxHR 444 10 bits video file, but in a MOV container. I don't know how this works internally, but maybe the color issue has something to do with the container metadata or something. I may try this if there's not another suggestion (transcoding this kind of files, as you know, takes time, so I'll wait for some ideas first).
NOTE: I tried to transcode the source file in the same way (DNxHR 444, 10 bits, etc) with Adobe Media Encoder 2020 and the result was the same, with colors messed up and FFprobe saying the video is BT709. Also tried transcoding to DNxHR HQX profile (10 bits), same result.
Any help would be greatly appreciated.
Thanks in advance.