[FFmpeg-devel] [PATCH] hevc: Add support for alternative transfer characterics SEI

[Vittorio Giovara]

On Sun, Jun 11, 2017 at 10:19 AM, James Almer <jamrial at gmail.com> wrote:
> On 6/9/2017 6:35 PM, Vittorio Giovara wrote:
>> Signed-off-by: Vittorio Giovara <vittorio.giovara at gmail.com>
>> ---
>>  libavcodec/hevc_sei.c | 9 +++++++++
>>  libavcodec/hevc_sei.h | 7 +++++++
>>  libavcodec/hevcdec.c  | 5 +++++
>>  3 files changed, 21 insertions(+)
>>
>> diff --git a/libavcodec/hevc_sei.c b/libavcodec/hevc_sei.c
>> index f0ad84f2f4..cd55d50212 100644
>> --- a/libavcodec/hevc_sei.c
>> +++ b/libavcodec/hevc_sei.c
>> @@ -265,6 +265,13 @@ static int decode_nal_sei_active_parameter_sets(HEVCSEIContext *s, GetBitContext
>>      return 0;
>>  }
>>
>> +static int decode_nal_sei_alternative_transfer(HEVCSEIAlternativeTransfer *s, GetBitContext *gb)
>> +{
>> +    s->present = 1;
>> +    s->preferred_transfer_characteristics = get_bits(gb, 8);
>> +    return 0;
>> +}
>> +
>>  static int decode_nal_sei_prefix(GetBitContext *gb, HEVCSEIContext *s, const HEVCParamSets *ps,
>>                                   int type, int size, void *logctx)
>>  {
>> @@ -285,6 +292,8 @@ static int decode_nal_sei_prefix(GetBitContext *gb, HEVCSEIContext *s, const HEV
>>          return decode_nal_sei_active_parameter_sets(s, gb, logctx);
>>      case HEVC_SEI_TYPE_USER_DATA_REGISTERED_ITU_T_T35:
>>          return decode_nal_sei_user_data_registered_itu_t_t35(s, gb, size);
>> +    case HEVC_SEI_TYPE_ALTERNATIVE_TRANSFER_CHARACTERISTICS:
>> +        return decode_nal_sei_alternative_transfer(&s->alternative_transfer, gb);
>>      default:
>>          av_log(logctx, AV_LOG_DEBUG, "Skipped PREFIX SEI %d\n", type);
>>          skip_bits_long(gb, 8 * size);
>> diff --git a/libavcodec/hevc_sei.h b/libavcodec/hevc_sei.h
>> index 5645e4f5de..1b522227ec 100644
>> --- a/libavcodec/hevc_sei.h
>> +++ b/libavcodec/hevc_sei.h
>> @@ -56,6 +56,7 @@ typedef enum {
>>      HEVC_SEI_TYPE_REGION_REFRESH_INFO                  = 134,
>>      HEVC_SEI_TYPE_MASTERING_DISPLAY_INFO               = 137,
>>      HEVC_SEI_TYPE_CONTENT_LIGHT_LEVEL_INFO             = 144,
>> +    HEVC_SEI_TYPE_ALTERNATIVE_TRANSFER_CHARACTERISTICS = 147,
>>  } HEVC_SEI_Type;
>>
>>  typedef struct HEVCSEIPictureHash {
>> @@ -100,6 +101,11 @@ typedef struct HEVCSEIContentLight {
>>      uint16_t max_pic_average_light_level;
>>  } HEVCSEIContentLight;
>>
>> +typedef struct HEVCSEIAlternativeTransfer {
>> +    int present;
>> +    int preferred_transfer_characteristics;
>> +} HEVCSEIAlternativeTransfer;
>> +
>>  typedef struct HEVCSEIContext {
>>      HEVCSEIPictureHash picture_hash;
>>      HEVCSEIFramePacking frame_packing;
>> @@ -109,6 +115,7 @@ typedef struct HEVCSEIContext {
>>      HEVCSEIMasteringDisplay mastering_display;
>>      HEVCSEIContentLight content_light;
>>      int active_seq_parameter_set_id;
>> +    HEVCSEIAlternativeTransfer alternative_transfer;
>>  } HEVCSEIContext;
>>
>>  struct HEVCParamSets;
>> diff --git a/libavcodec/hevcdec.c b/libavcodec/hevcdec.c
>> index 23a89345a8..955cc703cf 100644
>> --- a/libavcodec/hevcdec.c
>> +++ b/libavcodec/hevcdec.c
>> @@ -2684,6 +2684,11 @@ static int set_side_data(HEVCContext *s)
>>          s->avctx->properties |= FF_CODEC_PROPERTY_CLOSED_CAPTIONS;
>>      }
>>
>> +    if (s->sei.alternative_transfer.present &&
>> +        s->sei.alternative_transfer.preferred_transfer_characteristics != AVCOL_SPC_UNSPECIFIED) {
>
> Shouldn't this be AVCOL_TRC_UNSPECIFIED?

Right, typo, thanks.

> And validate it with av_color_transfer_name() instead, like it's done in
> hevc_ps.c with the VUI value.
>
>> +        s->avctx->color_trc = s->sei.alternative_transfer.preferred_transfer_characteristics;
>
> Did you make sure avctx->color_trc will not be overwritten by the VUI
> value after this? As in, what's called first, set_side_data() or
> export_stream_params()? And is it properly propagated in updates between
> thread contexts?

This is called after extradata decoding: from hevc_decode_frame(),
first hevc_decode_extradata() -> export_stream_param(), then
decode_nal_units() -> decode_nal_unit() -> hevc_frame_start() ->
set_side_data().
Regarding your second question, I suppose so, the sei check just above
modifies the global context in a similar way.

One thing I noticed is that I probably need to update the output frame
value besides the context, I'll send an updated version.
--
Vittorio