[FFmpeg-devel] [PATCH 1/3] avcodec/libaomenc: Add parameter for avif single image encoding

[James Almer]

On 2/17/2022 2:51 AM, Vignesh Venkatasubramanian wrote:
> Add a parameter to libaom-av1 encoder to enforce some of the single
> image constraints in the AV1 encoder. Setting this flag will limit
> the encoder to producing exactly one frame and the sequence header
> that is produced by the encoder will be conformant to the AVIF
> specification [1].
> 
> Part of Fixing Trac ticket #7621
> 
> [1] https://aomediacodec.github.io/av1-avif
> 
> Signed-off-by:: Vignesh Venkatasubramanian <vigneshv at google.com>
> ---
>   libavcodec/libaomenc.c | 14 ++++++++++++++
>   1 file changed, 14 insertions(+)
> 
> diff --git a/libavcodec/libaomenc.c b/libavcodec/libaomenc.c
> index 963cc1bcbc..0398060a2f 100644
> --- a/libavcodec/libaomenc.c
> +++ b/libavcodec/libaomenc.c
> @@ -99,6 +99,7 @@ typedef struct AOMEncoderContext {
>       int enable_restoration;
>       int usage;
>       int tune;
> +    int is_avif;
>       int enable_rect_partitions;
>       int enable_1to4_partitions;
>       int enable_ab_partitions;
> @@ -746,6 +747,18 @@ static av_cold int aom_init(AVCodecContext *avctx,
>       if (res < 0)
>           return res;
>   
> +    if (ctx->is_avif) {
> +        // Set the maximum number of frames to 1. This will let libaom set
> +        // still_picture and reduced_still_picture_header to 1 in the Sequence
> +        // Header as required by AVIF still images.
> +        enccfg.g_limit = 1;

How will libaom react if you feed it more than 1 frame? Will it reject 
the input and return an error?

> +        // Reduce memory usage for still images.
> +        enccfg.g_lag_in_frames = 0;
> +        // All frames will be key frames.
> +        enccfg.kf_max_dist = 0;
> +        enccfg.kf_mode = AOM_KF_DISABLED;
> +    }
> +
>       /* Construct Encoder Context */
>       res = aom_codec_enc_init(&ctx->encoder, iface, &enccfg, flags);
>       if (res != AOM_CODEC_OK) {
> @@ -1290,6 +1303,7 @@ static const AVOption options[] = {
>       { "psnr",            NULL,         0, AV_OPT_TYPE_CONST, {.i64 = AOM_TUNE_PSNR}, 0, 0, VE, "tune"},
>       { "ssim",            NULL,         0, AV_OPT_TYPE_CONST, {.i64 = AOM_TUNE_SSIM}, 0, 0, VE, "tune"},
>       FF_AV1_PROFILE_OPTS
> +    { "avif-image", "Encode in single frame mode for still AVIF images.", OFFSET(is_avif), AV_OPT_TYPE_BOOL, {.i64 = -1}, -1, 1, VE },

Judging by the above settings, this option should be called 
"still_picture". The description can mention it's required for AVIF output.

An alternative is to add new AVProfiles, giving them values outside the 
range that would be written into the sequence header. Like 
FF_PROFILE_AV1_MAIN_STILL_PICTURE with a value of 0x80, 
FF_PROFILE_AV1_HIGH_STILL_PICTURE with a value of 0x81, etc.
Then you just look for avctx->profile & 0x80 to know it's a still picture.

Of course considerations would need to be done to avoid problems, like 
the line "enccfg.g_profile = avctx->profile;" being changed to 
"enccfg.g_profile = avctx->profile & 0x3;" (We can define a constant for 
that 0x3 mask called profile_bits or whatever).

>       { "enable-rect-partitions", "Enable rectangular partitions", OFFSET(enable_rect_partitions), AV_OPT_TYPE_BOOL, {.i64 = -1}, -1, 1, VE},
>       { "enable-1to4-partitions", "Enable 1:4/4:1 partitions",     OFFSET(enable_1to4_partitions), AV_OPT_TYPE_BOOL, {.i64 = -1}, -1, 1, VE},
>       { "enable-ab-partitions",   "Enable ab shape partitions",    OFFSET(enable_ab_partitions),   AV_OPT_TYPE_BOOL, {.i64 = -1}, -1, 1, VE},