Implement support for encoding HDR10+ from JSON metadata file

[quietvoid]

The goal of this PR is to allow users to encode HDR video with HDR10+ metadata, in compliance with the AV1 HDR10+ specification.

In the current state, the changes make it possible to encode HDR10+ for both CLI and Rust API usage.

From CLI, the metadata is expected to be passed as a JSON metadata file, which follows the same format as encoders like x265. The metadata parsing/encoding is done through the hdr10plus crate. The library is probably less than ideal but it should be fine to parse things at init.

The CLI opt can be either --hdr10plus-json or --dhdr10-info (used by x265)

From Rust, the metadata must be encoded into the final T.35 and provided for the frames that require it.

Output bitstream validated with the AOM AV1 HDR10+ Validator.

Starting this as a draft, as some things need to be improved;

• [ ] Behaviour when some parsed metadata is invalid.
  • Either the encoder has to abort, or keeps going with an incomplete map of frame<->metadata.
• [x] Need to make the ST2094-40 prefix const.
• [x] Improve comments.
• ... probably more, feedback welcome.

[quietvoid]

As is, this is probably ready for review. I'm just not sure how the CLI encoder should behave when the JSON metadata partially fails to parse.

Also, hopefully it also works in rav1e-ch, as long as the frame numbers are continuous across GOPs. edit: rav1e-ch seems to work just fine.

[quietvoid]

It does appear that the current T.35 insertion code skips over some frames with show_frame=0 and show_existing_frame=1. At least compared to the reference stream in https://github.com/AOMediaCodec/av1-hdr10plus/wiki, which has the metadata OBUs for every frame.

Though the spec (draft) does say

HDR10+ Metadata OBUs are not provided when show_frame=0.

So I'm not sure if the ref sample is correct. cc @tdaede

There is also an updated figure of the structure: https://raw.githubusercontent.com/AOMediaCodec/av1-hdr10plus/0e2a5d82ec9b7c24e3c4bf84c31557eb99563dd1/obu_tu.png

[codecov-commenter]

Codecov Report

Patch coverage: 40.78% and project coverage change: -1.78% :warning:

Comparison is base (184bf4b) 88.10% compared to head (ee4d6d1) 86.32%.

:exclamation: Current head ee4d6d1 differs from pull request most recent head 82dd0df. Consider uploading reports for the commit 82dd0df to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #3000      +/-   ##
==========================================
- Coverage   88.10%   86.32%   -1.78%     
==========================================
  Files          85       89       +4     
  Lines       33596    34122     +526     
==========================================
- Hits        29599    29457     -142     
- Misses       3997     4665     +668     

Files Changed	Coverage Δ
src/api/util.rs	59.45% <ø> (+0.79%)	:arrow_up:
src/bin/common.rs	51.37% <6.89%> (-1.68%)	:arrow_down:
src/api/internal.rs	96.21% <59.09%> (-1.57%)	:arrow_down:
src/api/config/encoder.rs	89.36% <100.00%> (ø)
src/api/test.rs	99.23% <100.00%> (-0.38%)	:arrow_down:

... and 64 files with indirect coverage changes

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[quietvoid]

As the specification seems to be in a final state, I've reworked things to work properly for the OBU placement. https://aomediacodec.github.io/av1-hdr10plus/#placement

The T.35 metadata is added at the end when present in the EncoderConfig hdr10plus_payloads. And only if the existing frame T.35 metadata doesn't already contain one for HDR10+ metadata.

[lu-zero]

clippy seems unhappy, I let @tdaede confirm the actual specification details.

[quietvoid]

Doh. I was running clippy with --all-features and it was failing.

[quietvoid]

It was pointed out on the IRC that the payloads should probably be handled on the CLI side only, as the API EncoderConfig already has T.35 metadata support. However the current T.35 API implementation has issues with missing metadata that I've yet to resolve, so I'll update this when the issues are worked out.