Re: [FFmpeg-devel] [PATCH v4 1/2] avcodec/libvpxenc: add VP9 temporal scalability encoding option
HI,
On Mon, Jan 13, 2020 at 4:44 PM James Zern
wrote:
> Hi,
>
> On Fri, Jan 10, 2020 at 9:59 AM Wonkap Jang
> wrote:
> >
> > This commit reuses the configuration options for VP8 that enables
> > temporal scalability for VP9. It also adds a way to enable three
> > preset temporal structures (refer to the documentation for more
> > detail) that can be used in offline encoding.
> > ---
> > doc/encoders.texi | 18 ++-
> > libavcodec/libvpxenc.c | 251 +
> > 2 files changed, 243 insertions(+), 26 deletions(-)
> >
> > [...]
> > #endif
> > @@ -223,8 +231,16 @@ static av_cold void dump_enc_cfg(AVCodecContext
> *avctx,
> > " %*s%u\n", width, "ts_number_layers:",
> cfg->ts_number_layers);
> > av_log(avctx, level,
> > "\n %*s", width, "ts_target_bitrate:");
>
> This should move to the vp8 branch and a new heading added for vp9's
> layer_target_bitrate.
>
> > [...]
> > -static int vp8_ts_param_parse(struct vpx_codec_enc_cfg *enccfg, char
> *key, char *value)
> > +static void set_temporal_layer_pattern(int layering_mode,
> > + vpx_codec_enc_cfg_t *cfg,
> > + int *layer_flags,
> > + int *flag_periodicity)
> > +{
> > +switch (layering_mode) {
> > +case 2: {
> > +/**
> > + * 2-layers, 2-frame period.
> > + */
> > +int ids[2] = { 0, 1 };
>
> these can be static const.
>
[WJ] OK.
>
> > +cfg->ts_periodicity = 2;
> > +*flag_periodicity = 2;
> > +cfg->ts_number_layers = 2;
> > +cfg->ts_rate_decimator[0] = 2;
> > +cfg->ts_rate_decimator[1] = 1;
> > +memcpy(cfg->ts_layer_id, ids, sizeof(ids));
> > +
> > +layer_flags[0] =
> > + VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
> > + VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
> > +
>
> this line can be removed.
>
[WJ] OK.
>
> > +layer_flags[1] =
> > +VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_GF |
> > +VP8_EFLAG_NO_UPD_LAST |
> > +VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_REF_GF;
> > +break;
> > +}
> > +case 3: {
> > +/**
> > + * 3-layers structure with one reference frame.
> > + * This works same as temporal_layering_mode 3.
> > + *
> > + * 3-layers, 4-frame period.
> > + */
> > +int ids[4] = { 0, 2, 1, 2 };
> > +cfg->ts_periodicity = 4;
> > +*flag_periodicity = 4;
> > +cfg->ts_number_layers = 3;
> > +cfg->ts_rate_decimator[0] = 4;
> > +cfg->ts_rate_decimator[1] = 2;
> > +cfg->ts_rate_decimator[2] = 1;
> > +memcpy(cfg->ts_layer_id, ids, sizeof(ids));
> > +
> > +/**
> > + * 0=L, 1=GF, 2=ARF,
>
> what about [3]?
>
[WJ] decimator is for indicating the framerate decimation per temporal
layer. There are three temporal layers in this case, no need for [3].
>
> > + * Intra-layer prediction disabled.
> > + */
> > +layer_flags[0] =
> > +VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
> > +VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
> > +layer_flags[2] =
>
> can you reorder this to be in the natural 0...3 order?
>
[WJ] this is in the order of temporal layer id, just like it is written in
vpx_temporal_svc_encoder in libvpx/examples, but I'll make the change as
suggested.
>
> > [...]
> > +/**
> > + * 0=L, 1=GF, 2=ARF, Intra-layer prediction disabled.
> > + */
> > +layer_flags[0] =
> > +VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
> > +VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
> > +layer_flags[2] =
>
> same comment here about order.
>
> > [...]
> > +
> > +#if (VPX_ENCODER_ABI_VERSION >= 12) && CONFIG_LIBVPX_VP9_ENCODER
> > +enccfg->temporal_layering_mode = 1; // only bypass mode is being
> supported for now.
>
> ...mode is supported...
>
[WJ] gotit.
>
> > +enccfg->ss_number_layers = 1; // making sure the spatial
> scalability is off. Will support this in future.
>
> this can be TODO: add spatial scalability support.
[WJ] gotit.
>
> > +#endif
> > +if (ts_layering_mode) {
> > + // make sure the ts_layering_mode comes at the end of the
> ts_parameter string to ensure that
> > + // correct configuration is done.
> > + ctx->ts_layer_flags = av_malloc(sizeof(int) *
> VPX_TS_MAX_PERIODICITY);
>
> prefer sizeof(var).
>
[WJ] OK.
>
> > [...]
> >
> > -if(!avctx->bit_rate)
> > -if(avctx->rc_max_rate || avctx->rc_buffer_size ||
> avctx->rc_initial_buffer_occupancy) {
> > +if (!avctx->bit_rate)
> > +if (avctx->rc_max_rate || avctx->rc_buffer_size ||
> avctx->rc_initial_buffer_occupancy) {
>
> this is unrelated to the change, please remove it.
>
[WJ] ok.
>
> > [...]
> > +#if VPX_ENCODER_ABI_VERSION >= 12
> > +codecctl_int(avctx,
Re: [FFmpeg-devel] [PATCH v4 1/2] avcodec/libvpxenc: add VP9 temporal scalability encoding option
Hi,
On Fri, Jan 10, 2020 at 9:59 AM Wonkap Jang
wrote:
>
> This commit reuses the configuration options for VP8 that enables
> temporal scalability for VP9. It also adds a way to enable three
> preset temporal structures (refer to the documentation for more
> detail) that can be used in offline encoding.
> ---
> doc/encoders.texi | 18 ++-
> libavcodec/libvpxenc.c | 251 +
> 2 files changed, 243 insertions(+), 26 deletions(-)
>
> [...]
> #endif
> @@ -223,8 +231,16 @@ static av_cold void dump_enc_cfg(AVCodecContext *avctx,
> " %*s%u\n", width, "ts_number_layers:", cfg->ts_number_layers);
> av_log(avctx, level,
> "\n %*s", width, "ts_target_bitrate:");
This should move to the vp8 branch and a new heading added for vp9's
layer_target_bitrate.
> [...]
> -static int vp8_ts_param_parse(struct vpx_codec_enc_cfg *enccfg, char *key,
> char *value)
> +static void set_temporal_layer_pattern(int layering_mode,
> + vpx_codec_enc_cfg_t *cfg,
> + int *layer_flags,
> + int *flag_periodicity)
> +{
> +switch (layering_mode) {
> +case 2: {
> +/**
> + * 2-layers, 2-frame period.
> + */
> +int ids[2] = { 0, 1 };
these can be static const.
> +cfg->ts_periodicity = 2;
> +*flag_periodicity = 2;
> +cfg->ts_number_layers = 2;
> +cfg->ts_rate_decimator[0] = 2;
> +cfg->ts_rate_decimator[1] = 1;
> +memcpy(cfg->ts_layer_id, ids, sizeof(ids));
> +
> +layer_flags[0] =
> + VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
> + VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
> +
this line can be removed.
> +layer_flags[1] =
> +VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_GF |
> +VP8_EFLAG_NO_UPD_LAST |
> +VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_REF_GF;
> +break;
> +}
> +case 3: {
> +/**
> + * 3-layers structure with one reference frame.
> + * This works same as temporal_layering_mode 3.
> + *
> + * 3-layers, 4-frame period.
> + */
> +int ids[4] = { 0, 2, 1, 2 };
> +cfg->ts_periodicity = 4;
> +*flag_periodicity = 4;
> +cfg->ts_number_layers = 3;
> +cfg->ts_rate_decimator[0] = 4;
> +cfg->ts_rate_decimator[1] = 2;
> +cfg->ts_rate_decimator[2] = 1;
> +memcpy(cfg->ts_layer_id, ids, sizeof(ids));
> +
> +/**
> + * 0=L, 1=GF, 2=ARF,
what about [3]?
> + * Intra-layer prediction disabled.
> + */
> +layer_flags[0] =
> +VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
> +VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
> +layer_flags[2] =
can you reorder this to be in the natural 0...3 order?
> [...]
> +/**
> + * 0=L, 1=GF, 2=ARF, Intra-layer prediction disabled.
> + */
> +layer_flags[0] =
> +VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
> +VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
> +layer_flags[2] =
same comment here about order.
> [...]
> +
> +#if (VPX_ENCODER_ABI_VERSION >= 12) && CONFIG_LIBVPX_VP9_ENCODER
> +enccfg->temporal_layering_mode = 1; // only bypass mode is being
> supported for now.
...mode is supported...
> +enccfg->ss_number_layers = 1; // making sure the spatial scalability is
> off. Will support this in future.
this can be TODO: add spatial scalability support.
> +#endif
> +if (ts_layering_mode) {
> + // make sure the ts_layering_mode comes at the end of the ts_parameter
> string to ensure that
> + // correct configuration is done.
> + ctx->ts_layer_flags = av_malloc(sizeof(int) * VPX_TS_MAX_PERIODICITY);
prefer sizeof(var).
> [...]
>
> -if(!avctx->bit_rate)
> -if(avctx->rc_max_rate || avctx->rc_buffer_size ||
> avctx->rc_initial_buffer_occupancy) {
> +if (!avctx->bit_rate)
> +if (avctx->rc_max_rate || avctx->rc_buffer_size ||
> avctx->rc_initial_buffer_occupancy) {
this is unrelated to the change, please remove it.
> [...]
> +#if VPX_ENCODER_ABI_VERSION >= 12
> +codecctl_int(avctx, VP9E_SET_SVC, 1);
> +codecctl_intp(avctx, VP9E_SET_SVC_PARAMETERS, (int *) _params);
drop the space after the cast.
> [...]
> +#if CONFIG_LIBVPX_VP9_ENCODER && VPX_ENCODER_ABI_VERSION >= 12
> +else if (avctx->codec_id == AV_CODEC_ID_VP9) {
> +codecctl_intp(avctx, VP9E_SET_SVC_LAYER_ID, (int *) _id);
drop the space after the cast.
___
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
Re: [FFmpeg-devel] [PATCH v4 1/2] avcodec/libvpxenc: add VP9 temporal scalability encoding option
On Fri, Jan 10, 2020 at 9:59 AM Wonkap Jang wrote:
> This commit reuses the configuration options for VP8 that enables
> temporal scalability for VP9. It also adds a way to enable three
> preset temporal structures (refer to the documentation for more
> detail) that can be used in offline encoding.
> ---
> doc/encoders.texi | 18 ++-
> libavcodec/libvpxenc.c | 251 +
> 2 files changed, 243 insertions(+), 26 deletions(-)
>
> diff --git a/doc/encoders.texi b/doc/encoders.texi
> index 61e674cf96..88429aed4c 100644
> --- a/doc/encoders.texi
> +++ b/doc/encoders.texi
> @@ -1885,8 +1885,6 @@ Enable error resiliency features.
> Increase sharpness at the expense of lower PSNR.
> The valid range is [0, 7].
>
> -@item VP8-specific options
> -@table @option
> @item ts-parameters
> Sets the temporal scalability configuration using a :-separated list of
> key=value pairs. For example, to specify temporal scalability parameters
> @@ -1894,7 +1892,7 @@ with @code{ffmpeg}:
> @example
> ffmpeg -i INPUT -c:v libvpx -ts-parameters ts_number_layers=3:\
> ts_target_bitrate=250,500,1000:ts_rate_decimator=4,2,1:\
> -ts_periodicity=4:ts_layer_id=0,2,1,2 OUTPUT
> +ts_periodicity=4:ts_layer_id=0,2,1,2:ts_layering_mode=3 OUTPUT
> @end example
> Below is a brief explanation of each of the parameters, please
> refer to @code{struct vpx_codec_enc_cfg} in @code{vpx/vpx_encoder.h} for
> more
> @@ -1911,6 +1909,20 @@ Frame rate decimation factor for each temporal
> layer.
> Length of the sequence defining frame temporal layer membership.
> @item ts_layer_id
> Template defining the membership of frames to temporal layers.
> +@item ts_layering_mode
> +(optional) Selecting the temporal structure from a set of pre-defined
> temporal layering modes.
> +Currently supports the following options.
> +@table @option
> +@item 0
> +No temporal layering flags are provided internally,
> +relies on flags being passed in using metadata in AVFrame.
> +@item 2
> +Two temporal layers. 0-1...
> +@item 3
> +Three temporal layers. 0-2-1-2...; with single reference frame.
> +@item 4
> +Same as option "3", except there is a dependency between
> +the two temporal layer 2 frames within the temporal period.
> @end table
> @end table
>
> diff --git a/libavcodec/libvpxenc.c b/libavcodec/libvpxenc.c
> index 0b8a070304..14cc1e7158 100644
> --- a/libavcodec/libvpxenc.c
> +++ b/libavcodec/libvpxenc.c
> @@ -100,7 +100,9 @@ typedef struct VPxEncoderContext {
> int rc_undershoot_pct;
> int rc_overshoot_pct;
>
> -AVDictionary *vp8_ts_parameters;
> +AVDictionary *vpx_ts_parameters;
> +int *ts_layer_flags;
> +int current_temporal_idx;
>
> // VP9-only
> int lossless;
> @@ -137,6 +139,7 @@ static const char *const ctlidstr[] = {
> [VP8E_SET_CQ_LEVEL] = "VP8E_SET_CQ_LEVEL",
> [VP8E_SET_MAX_INTRA_BITRATE_PCT] = "VP8E_SET_MAX_INTRA_BITRATE_PCT",
> [VP8E_SET_SHARPNESS] = "VP8E_SET_SHARPNESS",
> +[VP8E_SET_TEMPORAL_LAYER_ID] = "VP8E_SET_TEMPORAL_LAYER_ID",
> #if CONFIG_LIBVPX_VP9_ENCODER
> [VP9E_SET_LOSSLESS]= "VP9E_SET_LOSSLESS",
> [VP9E_SET_TILE_COLUMNS]= "VP9E_SET_TILE_COLUMNS",
> @@ -144,6 +147,11 @@ static const char *const ctlidstr[] = {
> [VP9E_SET_FRAME_PARALLEL_DECODING] =
> "VP9E_SET_FRAME_PARALLEL_DECODING",
> [VP9E_SET_AQ_MODE] = "VP9E_SET_AQ_MODE",
> [VP9E_SET_COLOR_SPACE] = "VP9E_SET_COLOR_SPACE",
> +[VP9E_SET_SVC_LAYER_ID]= "VP9E_SET_SVC_LAYER_ID",
> +#if VPX_ENCODER_ABI_VERSION >= 12
> +[VP9E_SET_SVC_PARAMETERS] = "VP9E_SET_SVC_PARAMETERS",
> +#endif
> +[VP9E_SET_SVC] = "VP9E_SET_SVC",
> #if VPX_ENCODER_ABI_VERSION >= 11
> [VP9E_SET_COLOR_RANGE] = "VP9E_SET_COLOR_RANGE",
> #endif
> @@ -223,8 +231,16 @@ static av_cold void dump_enc_cfg(AVCodecContext
> *avctx,
> " %*s%u\n", width, "ts_number_layers:",
> cfg->ts_number_layers);
> av_log(avctx, level,
> "\n %*s", width, "ts_target_bitrate:");
> -for (i = 0; i < VPX_TS_MAX_LAYERS; i++)
> -av_log(avctx, level, "%u ", cfg->ts_target_bitrate[i]);
> +if (avctx->codec_id == AV_CODEC_ID_VP8) {
> +for (i = 0; i < VPX_TS_MAX_LAYERS; i++)
> +av_log(avctx, level, "%u ", cfg->ts_target_bitrate[i]);
> +}
> +#if (VPX_ENCODER_ABI_VERSION >= 12) && CONFIG_LIBVPX_VP9_ENCODER
> +if (avctx->codec_id == AV_CODEC_ID_VP9) {
> +for (i = 0; i < VPX_TS_MAX_LAYERS; i++)
> +av_log(avctx, level, "%u ", cfg->layer_target_bitrate[i]);
> +}
> +#endif
> av_log(avctx, level, "\n");
> av_log(avctx, level,
> "\n %*s", width, "ts_rate_decimator:");
> @@ -346,6 +362,8 @@ static av_cold int vpx_free(AVCodecContext *avctx)
> }
> #endif
>
> +av_freep(>ts_layer_flags);
> +
> vpx_codec_destroy(>encoder);
> if (ctx->is_alpha)
[FFmpeg-devel] [PATCH v4 1/2] avcodec/libvpxenc: add VP9 temporal scalability encoding option
This commit reuses the configuration options for VP8 that enables
temporal scalability for VP9. It also adds a way to enable three
preset temporal structures (refer to the documentation for more
detail) that can be used in offline encoding.
---
doc/encoders.texi | 18 ++-
libavcodec/libvpxenc.c | 251 +
2 files changed, 243 insertions(+), 26 deletions(-)
diff --git a/doc/encoders.texi b/doc/encoders.texi
index 61e674cf96..88429aed4c 100644
--- a/doc/encoders.texi
+++ b/doc/encoders.texi
@@ -1885,8 +1885,6 @@ Enable error resiliency features.
Increase sharpness at the expense of lower PSNR.
The valid range is [0, 7].
-@item VP8-specific options
-@table @option
@item ts-parameters
Sets the temporal scalability configuration using a :-separated list of
key=value pairs. For example, to specify temporal scalability parameters
@@ -1894,7 +1892,7 @@ with @code{ffmpeg}:
@example
ffmpeg -i INPUT -c:v libvpx -ts-parameters ts_number_layers=3:\
ts_target_bitrate=250,500,1000:ts_rate_decimator=4,2,1:\
-ts_periodicity=4:ts_layer_id=0,2,1,2 OUTPUT
+ts_periodicity=4:ts_layer_id=0,2,1,2:ts_layering_mode=3 OUTPUT
@end example
Below is a brief explanation of each of the parameters, please
refer to @code{struct vpx_codec_enc_cfg} in @code{vpx/vpx_encoder.h} for more
@@ -1911,6 +1909,20 @@ Frame rate decimation factor for each temporal layer.
Length of the sequence defining frame temporal layer membership.
@item ts_layer_id
Template defining the membership of frames to temporal layers.
+@item ts_layering_mode
+(optional) Selecting the temporal structure from a set of pre-defined temporal
layering modes.
+Currently supports the following options.
+@table @option
+@item 0
+No temporal layering flags are provided internally,
+relies on flags being passed in using metadata in AVFrame.
+@item 2
+Two temporal layers. 0-1...
+@item 3
+Three temporal layers. 0-2-1-2...; with single reference frame.
+@item 4
+Same as option "3", except there is a dependency between
+the two temporal layer 2 frames within the temporal period.
@end table
@end table
diff --git a/libavcodec/libvpxenc.c b/libavcodec/libvpxenc.c
index 0b8a070304..14cc1e7158 100644
--- a/libavcodec/libvpxenc.c
+++ b/libavcodec/libvpxenc.c
@@ -100,7 +100,9 @@ typedef struct VPxEncoderContext {
int rc_undershoot_pct;
int rc_overshoot_pct;
-AVDictionary *vp8_ts_parameters;
+AVDictionary *vpx_ts_parameters;
+int *ts_layer_flags;
+int current_temporal_idx;
// VP9-only
int lossless;
@@ -137,6 +139,7 @@ static const char *const ctlidstr[] = {
[VP8E_SET_CQ_LEVEL] = "VP8E_SET_CQ_LEVEL",
[VP8E_SET_MAX_INTRA_BITRATE_PCT] = "VP8E_SET_MAX_INTRA_BITRATE_PCT",
[VP8E_SET_SHARPNESS] = "VP8E_SET_SHARPNESS",
+[VP8E_SET_TEMPORAL_LAYER_ID] = "VP8E_SET_TEMPORAL_LAYER_ID",
#if CONFIG_LIBVPX_VP9_ENCODER
[VP9E_SET_LOSSLESS]= "VP9E_SET_LOSSLESS",
[VP9E_SET_TILE_COLUMNS]= "VP9E_SET_TILE_COLUMNS",
@@ -144,6 +147,11 @@ static const char *const ctlidstr[] = {
[VP9E_SET_FRAME_PARALLEL_DECODING] = "VP9E_SET_FRAME_PARALLEL_DECODING",
[VP9E_SET_AQ_MODE] = "VP9E_SET_AQ_MODE",
[VP9E_SET_COLOR_SPACE] = "VP9E_SET_COLOR_SPACE",
+[VP9E_SET_SVC_LAYER_ID]= "VP9E_SET_SVC_LAYER_ID",
+#if VPX_ENCODER_ABI_VERSION >= 12
+[VP9E_SET_SVC_PARAMETERS] = "VP9E_SET_SVC_PARAMETERS",
+#endif
+[VP9E_SET_SVC] = "VP9E_SET_SVC",
#if VPX_ENCODER_ABI_VERSION >= 11
[VP9E_SET_COLOR_RANGE] = "VP9E_SET_COLOR_RANGE",
#endif
@@ -223,8 +231,16 @@ static av_cold void dump_enc_cfg(AVCodecContext *avctx,
" %*s%u\n", width, "ts_number_layers:", cfg->ts_number_layers);
av_log(avctx, level,
"\n %*s", width, "ts_target_bitrate:");
-for (i = 0; i < VPX_TS_MAX_LAYERS; i++)
-av_log(avctx, level, "%u ", cfg->ts_target_bitrate[i]);
+if (avctx->codec_id == AV_CODEC_ID_VP8) {
+for (i = 0; i < VPX_TS_MAX_LAYERS; i++)
+av_log(avctx, level, "%u ", cfg->ts_target_bitrate[i]);
+}
+#if (VPX_ENCODER_ABI_VERSION >= 12) && CONFIG_LIBVPX_VP9_ENCODER
+if (avctx->codec_id == AV_CODEC_ID_VP9) {
+for (i = 0; i < VPX_TS_MAX_LAYERS; i++)
+av_log(avctx, level, "%u ", cfg->layer_target_bitrate[i]);
+}
+#endif
av_log(avctx, level, "\n");
av_log(avctx, level,
"\n %*s", width, "ts_rate_decimator:");
@@ -346,6 +362,8 @@ static av_cold int vpx_free(AVCodecContext *avctx)
}
#endif
+av_freep(>ts_layer_flags);
+
vpx_codec_destroy(>encoder);
if (ctx->is_alpha) {
vpx_codec_destroy(>encoder_alpha);
@@ -370,23 +388,154 @@ static void vp8_ts_parse_int_array(int *dest, char
*value, size_t value_len, int
}
}
-static int vp8_ts_param_parse(struct vpx_codec_enc_cfg *enccfg, char *key,
char *value)
+static void
