[FFmpeg-devel] [PATCH v2] avfilter/scale*: add option reset_sar
Gyan Doshi
ffmpeg at gyani.pro
Mon Feb 10 12:56:17 EET 2025
On 2025-02-09 03:41 pm, Gyan Doshi wrote:
> For anamorphic videos, enabling this option leads to adjustment of
> output dimensions to obtain square pixels when the user requests
> proportional scaling through either of the w/h expressions or
> force_original_aspect_ratio.
>
> Output SAR is always reset to 1.
>
> Option added to scale, scale_cuda, scale_npp & scale_vaapi.
>
> libplacebo already has a similar option with different semantics,
> scale_vt and scale_vulkan don't implement force_oar, so for these
> three filters, I've made minimal changes needed to not break building
> or change output.
> ---
> v2:
> Rewrote doc option description
> Added doc examples
> w_adj init nit resolved
Pushed as a28dc06869fe1f98c07e42f9b0a411d2744ff7d7
Regards,
Gyan
>
> doc/filters.texi | 36 +++++++++++++++++++++++++++++++++--
> libavfilter/scale_eval.c | 13 +++++++------
> libavfilter/scale_eval.h | 5 ++++-
> libavfilter/vf_libplacebo.c | 2 +-
> libavfilter/vf_scale.c | 13 +++++++++++--
> libavfilter/vf_scale_cuda.c | 25 ++++++++++++++++++------
> libavfilter/vf_scale_npp.c | 13 +++++++++++--
> libavfilter/vf_scale_vaapi.c | 13 +++++++++++--
> libavfilter/vf_scale_vt.c | 2 +-
> libavfilter/vf_scale_vulkan.c | 2 +-
> 10 files changed, 100 insertions(+), 24 deletions(-)
>
> diff --git a/doc/filters.texi b/doc/filters.texi
> index a14c7e7e77..bf9a60712f 100644
> --- a/doc/filters.texi
> +++ b/doc/filters.texi
> @@ -21285,6 +21285,14 @@ This option can be handy if you need to have a video fit within or exceed
> a defined resolution using @option{force_original_aspect_ratio} but also have
> encoder restrictions on width or height divisibility.
>
> + at item reset_sar
> +Enabling this option leads to the output SAR being reset to 1.
> +Additionally, if the user requests proportional scaling either
> +through the width or height expressions, e.g. @code{w=-4:h=360} or @code{w=iw/2:h=-1}
> +or by enabling @code{force_original_aspect_ratio}, then the input DAR is taken into
> +account and the output is scaled to produce square pixels.
> +Default is false.
> +
> @end table
>
> The values of the @option{w} and @option{h} options are expressions
> @@ -21445,10 +21453,28 @@ scale='trunc(ih*dar):ih',setsar=1/1
> @end example
>
> @item
> -Make pixels square by combining scale and setsar,
> +Make pixels square using reset_sar,
> making sure the resulting resolution is even (required by some codecs):
> @example
> -scale='trunc(ih*dar/2)*2:trunc(ih/2)*2',setsar=1/1
> +scale='-2:ih-mod(ih,2):reset_sar=1'
> + at end example
> +
> + at item
> +Scale to target exactly, however reset SAR to 1:
> + at example
> +scale='400:300:reset_sar=1'
> + at end example
> +
> + at item
> +Scale to even dimensions that fit within 400x300, preserving input SAR:
> + at example
> +scale='400:300:force_original_aspect_ratio=decrease:force_divisible_by=2'
> + at end example
> +
> + at item
> +Scale to produce square pixels with even dimensions that fit within 400x300:
> + at example
> +scale='400:300:force_original_aspect_ratio=decrease:force_divisible_by=2:reset_sar=1'
> @end example
>
> @item
> @@ -21538,6 +21564,9 @@ Affects the curves of the bicubic algorithm.
> @item force_divisible_by
> Work the same as the identical @ref{scale} filter options.
>
> + at item reset_sar
> +Works the same as the identical @ref{scale} filter option.
> +
> @end table
>
> @subsection Examples
> @@ -21641,6 +21670,9 @@ This option can be handy if you need to have a video fit within or exceed
> a defined resolution using @option{force_original_aspect_ratio} but also have
> encoder restrictions on width or height divisibility.
>
> + at item reset_sar
> +Works the same as the identical @ref{scale} filter option.
> +
> @item eval
> Specify when to evaluate @var{width} and @var{height} expression. It accepts the following values:
>
> diff --git a/libavfilter/scale_eval.c b/libavfilter/scale_eval.c
> index dc8d522b1e..53f5e22b0e 100644
> --- a/libavfilter/scale_eval.c
> +++ b/libavfilter/scale_eval.c
> @@ -112,7 +112,8 @@ fail:
>
> int ff_scale_adjust_dimensions(AVFilterLink *inlink,
> int *ret_w, int *ret_h,
> - int force_original_aspect_ratio, int force_divisible_by)
> + int force_original_aspect_ratio, int force_divisible_by,
> + double w_adj)
> {
> int64_t w, h;
> int factor_w, factor_h;
> @@ -132,7 +133,7 @@ int ff_scale_adjust_dimensions(AVFilterLink *inlink,
> }
>
> if (w < 0 && h < 0) {
> - w = inlink->w;
> + w = inlink->w * w_adj;
> h = inlink->h;
> }
>
> @@ -140,18 +141,18 @@ int ff_scale_adjust_dimensions(AVFilterLink *inlink,
> * earlier. If no factor was set, nothing will happen as the default
> * factor is 1 */
> if (w < 0)
> - w = av_rescale(h, inlink->w, inlink->h * factor_w) * factor_w;
> + w = av_rescale(h, inlink->w * w_adj, inlink->h * factor_w) * factor_w;
> if (h < 0)
> - h = av_rescale(w, inlink->h, inlink->w * factor_h) * factor_h;
> + h = av_rescale(w, inlink->h, inlink->w * w_adj * factor_h) * factor_h;
>
> /* Note that force_original_aspect_ratio may overwrite the previous set
> * dimensions so that it is not divisible by the set factors anymore
> * unless force_divisible_by is defined as well */
> if (force_original_aspect_ratio) {
> // Including force_divisible_by here rounds to the nearest multiple of it.
> - int64_t tmp_w = av_rescale(h, inlink->w, inlink->h * (int64_t)force_divisible_by)
> + int64_t tmp_w = av_rescale(h, inlink->w * w_adj, inlink->h * (int64_t)force_divisible_by)
> * force_divisible_by;
> - int64_t tmp_h = av_rescale(w, inlink->h, inlink->w * (int64_t)force_divisible_by)
> + int64_t tmp_h = av_rescale(w, inlink->h, inlink->w * w_adj * (int64_t)force_divisible_by)
> * force_divisible_by;
>
> if (force_original_aspect_ratio == 1) {
> diff --git a/libavfilter/scale_eval.h b/libavfilter/scale_eval.h
> index b489528404..6cab623e1c 100644
> --- a/libavfilter/scale_eval.h
> +++ b/libavfilter/scale_eval.h
> @@ -40,10 +40,13 @@ int ff_scale_eval_dimensions(void *ctx,
> * or both of the evaluated values are of the form '-n' or if
> * force_original_aspect_ratio is set. force_divisible_by is used only when
> * force_original_aspect_ratio is set and must be at least 1.
> + * w_adj is the input SAR when the output dimensions are intended to be square
> + * pixels, else should be 1.
> *
> * Returns negative error code on error or non negative on success
> */
> int ff_scale_adjust_dimensions(AVFilterLink *inlink,
> int *ret_w, int *ret_h,
> - int force_original_aspect_ratio, int force_divisible_by);
> + int force_original_aspect_ratio, int force_divisible_by,
> + double w_adj);
> #endif
> diff --git a/libavfilter/vf_libplacebo.c b/libavfilter/vf_libplacebo.c
> index c884c621c8..e1c6629f6d 100644
> --- a/libavfilter/vf_libplacebo.c
> +++ b/libavfilter/vf_libplacebo.c
> @@ -1201,7 +1201,7 @@ static int libplacebo_config_output(AVFilterLink *outlink)
>
> ff_scale_adjust_dimensions(inlink, &outlink->w, &outlink->h,
> s->force_original_aspect_ratio,
> - s->force_divisible_by);
> + s->force_divisible_by, 1.f);
>
> if (s->normalize_sar || s->nb_inputs > 1) {
> /* SAR is normalized, or we have multiple inputs, set out to 1:1 */
> diff --git a/libavfilter/vf_scale.c b/libavfilter/vf_scale.c
> index 14ce1fbbd8..4b0b08d9d4 100644
> --- a/libavfilter/vf_scale.c
> +++ b/libavfilter/vf_scale.c
> @@ -176,6 +176,7 @@ typedef struct ScaleContext {
>
> int force_original_aspect_ratio;
> int force_divisible_by;
> + int reset_sar;
>
> int eval_mode; ///< expression evaluation mode
>
> @@ -645,6 +646,7 @@ static int config_props(AVFilterLink *outlink)
> outlink->src->inputs[0];
> ScaleContext *scale = ctx->priv;
> uint8_t *flags_val = NULL;
> + double w_adj = 1.0;
> int ret;
>
> if ((ret = scale_eval_dimensions(ctx)) < 0)
> @@ -653,9 +655,13 @@ static int config_props(AVFilterLink *outlink)
> outlink->w = scale->w;
> outlink->h = scale->h;
>
> + if (scale->reset_sar)
> + w_adj = IS_SCALE2REF(ctx) ? scale->var_values[VAR_S2R_MAIN_SAR] :
> + scale->var_values[VAR_SAR];
> +
> ret = ff_scale_adjust_dimensions(inlink, &outlink->w, &outlink->h,
> scale->force_original_aspect_ratio,
> - scale->force_divisible_by);
> + scale->force_divisible_by, w_adj);
>
> if (ret < 0)
> goto fail;
> @@ -668,7 +674,9 @@ static int config_props(AVFilterLink *outlink)
>
> /* TODO: make algorithm configurable */
>
> - if (inlink0->sample_aspect_ratio.num){
> + if (scale->reset_sar)
> + outlink->sample_aspect_ratio = (AVRational){1, 1};
> + else if (inlink0->sample_aspect_ratio.num){
> outlink->sample_aspect_ratio = av_mul_q((AVRational){outlink->h * inlink0->w, outlink->w * inlink0->h}, inlink0->sample_aspect_ratio);
> } else
> outlink->sample_aspect_ratio = inlink0->sample_aspect_ratio;
> @@ -1175,6 +1183,7 @@ static const AVOption scale_options[] = {
> { "decrease", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = 1 }, 0, 0, FLAGS, .unit = "force_oar" },
> { "increase", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = 2 }, 0, 0, FLAGS, .unit = "force_oar" },
> { "force_divisible_by", "enforce that the output resolution is divisible by a defined integer when force_original_aspect_ratio is used", OFFSET(force_divisible_by), AV_OPT_TYPE_INT, { .i64 = 1}, 1, 256, FLAGS },
> + { "reset_sar", "scale output frame to have square pixels", OFFSET(reset_sar), AV_OPT_TYPE_BOOL, { .i64 = 0}, 0, 1, FLAGS },
> { "param0", "Scaler param 0", OFFSET(param[0]), AV_OPT_TYPE_DOUBLE, { .dbl = DBL_MAX }, -DBL_MAX, DBL_MAX, FLAGS },
> { "param1", "Scaler param 1", OFFSET(param[1]), AV_OPT_TYPE_DOUBLE, { .dbl = DBL_MAX }, -DBL_MAX, DBL_MAX, FLAGS },
> { "eval", "specify when to evaluate expressions", OFFSET(eval_mode), AV_OPT_TYPE_INT, {.i64 = EVAL_MODE_INIT}, 0, EVAL_MODE_NB-1, FLAGS, .unit = "eval" },
> diff --git a/libavfilter/vf_scale_cuda.c b/libavfilter/vf_scale_cuda.c
> index 5f9fd59118..de1883a1df 100644
> --- a/libavfilter/vf_scale_cuda.c
> +++ b/libavfilter/vf_scale_cuda.c
> @@ -96,6 +96,7 @@ typedef struct CUDAScaleContext {
>
> int force_original_aspect_ratio;
> int force_divisible_by;
> + int reset_sar;
>
> CUcontext cu_ctx;
> CUmodule cu_module;
> @@ -355,6 +356,7 @@ static av_cold int cudascale_config_props(AVFilterLink *outlink)
> AVHWFramesContext *frames_ctx;
> AVCUDADeviceContext *device_hwctx;
> int w, h;
> + double w_adj = 1.0;
> int ret;
>
> if ((ret = ff_scale_eval_dimensions(s,
> @@ -363,8 +365,12 @@ static av_cold int cudascale_config_props(AVFilterLink *outlink)
> &w, &h)) < 0)
> goto fail;
>
> + if (s->reset_sar)
> + w_adj = inlink->sample_aspect_ratio.num ?
> + (double)inlink->sample_aspect_ratio.num / inlink->sample_aspect_ratio.den : 1;
> +
> ff_scale_adjust_dimensions(inlink, &w, &h,
> - s->force_original_aspect_ratio, s->force_divisible_by);
> + s->force_original_aspect_ratio, s->force_divisible_by, w_adj);
>
> if (((int64_t)h * inlink->w) > INT_MAX ||
> ((int64_t)w * inlink->h) > INT_MAX)
> @@ -383,7 +389,9 @@ static av_cold int cudascale_config_props(AVFilterLink *outlink)
> s->hwctx = device_hwctx;
> s->cu_stream = s->hwctx->stream;
>
> - if (inlink->sample_aspect_ratio.num) {
> + if (s->reset_sar)
> + outlink->sample_aspect_ratio = (AVRational){1, 1};
> + else if (inlink->sample_aspect_ratio.num) {
> outlink->sample_aspect_ratio = av_mul_q((AVRational){outlink->h*inlink->w,
> outlink->w*inlink->h},
> inlink->sample_aspect_ratio);
> @@ -574,10 +582,14 @@ static int cudascale_filter_frame(AVFilterLink *link, AVFrame *in)
> if (ret < 0)
> goto fail;
>
> - av_reduce(&out->sample_aspect_ratio.num, &out->sample_aspect_ratio.den,
> - (int64_t)in->sample_aspect_ratio.num * outlink->h * link->w,
> - (int64_t)in->sample_aspect_ratio.den * outlink->w * link->h,
> - INT_MAX);
> + if (s->reset_sar) {
> + out->sample_aspect_ratio = (AVRational){1, 1};
> + } else {
> + av_reduce(&out->sample_aspect_ratio.num, &out->sample_aspect_ratio.den,
> + (int64_t)in->sample_aspect_ratio.num * outlink->h * link->w,
> + (int64_t)in->sample_aspect_ratio.den * outlink->w * link->h,
> + INT_MAX);
> + }
>
> av_frame_free(&in);
> return ff_filter_frame(outlink, out);
> @@ -614,6 +626,7 @@ static const AVOption options[] = {
> { "decrease", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = 1 }, 0, 0, FLAGS, .unit = "force_oar" },
> { "increase", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = 2 }, 0, 0, FLAGS, .unit = "force_oar" },
> { "force_divisible_by", "enforce that the output resolution is divisible by a defined integer when force_original_aspect_ratio is used", OFFSET(force_divisible_by), AV_OPT_TYPE_INT, { .i64 = 1 }, 1, 256, FLAGS },
> + { "reset_sar", "scale output frame to have square pixels", OFFSET(reset_sar), AV_OPT_TYPE_BOOL, { .i64 = 0}, 0, 1, FLAGS },
> { NULL },
> };
>
> diff --git a/libavfilter/vf_scale_npp.c b/libavfilter/vf_scale_npp.c
> index ec3267801b..5274158f1b 100644
> --- a/libavfilter/vf_scale_npp.c
> +++ b/libavfilter/vf_scale_npp.c
> @@ -160,6 +160,7 @@ typedef struct NPPScaleContext {
>
> int force_original_aspect_ratio;
> int force_divisible_by;
> + int reset_sar;
>
> int interp_algo;
>
> @@ -650,14 +651,19 @@ static int config_props(AVFilterLink *outlink)
> outlink->src->inputs[1] :
> outlink->src->inputs[0];
> NPPScaleContext *s = ctx->priv;
> + double w_adj = 1.0;
> int ret;
>
> if ((ret = nppscale_eval_dimensions(ctx)) < 0)
> goto fail;
>
> + if (s->reset_sar)
> + w_adj = IS_SCALE2REF(ctx) ? s->var_values[VAR_S2R_MAIN_SAR] :
> + s->var_values[VAR_SAR];
> +
> ff_scale_adjust_dimensions(inlink, &s->w, &s->h,
> s->force_original_aspect_ratio,
> - s->force_divisible_by);
> + s->force_divisible_by, w_adj);
>
> if (s->w > INT_MAX || s->h > INT_MAX ||
> (s->h * inlink->w) > INT_MAX ||
> @@ -674,7 +680,9 @@ static int config_props(AVFilterLink *outlink)
> av_log(ctx, AV_LOG_VERBOSE, "w:%d h:%d -> w:%d h:%d\n",
> inlink->w, inlink->h, outlink->w, outlink->h);
>
> - if (inlink->sample_aspect_ratio.num)
> + if (s->reset_sar)
> + outlink->sample_aspect_ratio = (AVRational){1, 1};
> + else if (inlink->sample_aspect_ratio.num)
> outlink->sample_aspect_ratio = av_mul_q((AVRational){outlink->h*inlink->w,
> outlink->w*inlink->h},
> inlink->sample_aspect_ratio);
> @@ -1019,6 +1027,7 @@ static const AVOption options[] = {
> { "decrease", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = 1 }, 0, 0, FLAGS, .unit = "force_oar" },
> { "increase", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = 2 }, 0, 0, FLAGS, .unit = "force_oar" },
> { "force_divisible_by", "enforce that the output resolution is divisible by a defined integer when force_original_aspect_ratio is used", OFFSET(force_divisible_by), AV_OPT_TYPE_INT, { .i64 = 1 }, 1, 256, FLAGS },
> + { "reset_sar", "scale output frame to have square pixels", OFFSET(reset_sar), AV_OPT_TYPE_BOOL, { .i64 = 0}, 0, 1, FLAGS },
> { "eval", "specify when to evaluate expressions", OFFSET(eval_mode), AV_OPT_TYPE_INT, { .i64 = EVAL_MODE_INIT }, 0, EVAL_MODE_NB-1, FLAGS, .unit = "eval" },
> { "init", "eval expressions once during initialization", 0, AV_OPT_TYPE_CONST, { .i64 = EVAL_MODE_INIT }, 0, 0, FLAGS, .unit = "eval" },
> { "frame", "eval expressions during initialization and per-frame", 0, AV_OPT_TYPE_CONST, { .i64 = EVAL_MODE_FRAME }, 0, 0, FLAGS, .unit = "eval" },
> diff --git a/libavfilter/vf_scale_vaapi.c b/libavfilter/vf_scale_vaapi.c
> index 0f2d617996..ab95d41e0e 100644
> --- a/libavfilter/vf_scale_vaapi.c
> +++ b/libavfilter/vf_scale_vaapi.c
> @@ -39,6 +39,7 @@ typedef struct ScaleVAAPIContext {
>
> int force_original_aspect_ratio;
> int force_divisible_by;
> + int reset_sar;
>
> char *colour_primaries_string;
> char *colour_transfer_string;
> @@ -73,6 +74,7 @@ static int scale_vaapi_config_output(AVFilterLink *outlink)
> AVFilterContext *avctx = outlink->src;
> VAAPIVPPContext *vpp_ctx = avctx->priv;
> ScaleVAAPIContext *ctx = avctx->priv;
> + double w_adj = 1.0;
> int err;
>
> if ((err = ff_scale_eval_dimensions(ctx,
> @@ -81,8 +83,12 @@ static int scale_vaapi_config_output(AVFilterLink *outlink)
> &vpp_ctx->output_width, &vpp_ctx->output_height)) < 0)
> return err;
>
> + if (ctx->reset_sar)
> + w_adj = inlink->sample_aspect_ratio.num ?
> + (double)inlink->sample_aspect_ratio.num / inlink->sample_aspect_ratio.den : 1;
> +
> ff_scale_adjust_dimensions(inlink, &vpp_ctx->output_width, &vpp_ctx->output_height,
> - ctx->force_original_aspect_ratio, ctx->force_divisible_by);
> + ctx->force_original_aspect_ratio, ctx->force_divisible_by, w_adj);
>
> if (inlink->w == vpp_ctx->output_width && inlink->h == vpp_ctx->output_height &&
> (vpp_ctx->input_frames->sw_format == vpp_ctx->output_format ||
> @@ -98,7 +104,9 @@ static int scale_vaapi_config_output(AVFilterLink *outlink)
> if (err < 0)
> return err;
>
> - if (inlink->sample_aspect_ratio.num)
> + if (ctx->reset_sar)
> + outlink->sample_aspect_ratio = (AVRational){1, 1};
> + else if (inlink->sample_aspect_ratio.num)
> outlink->sample_aspect_ratio = av_mul_q((AVRational){outlink->h * inlink->w, outlink->w * inlink->h}, inlink->sample_aspect_ratio);
> else
> outlink->sample_aspect_ratio = inlink->sample_aspect_ratio;
> @@ -274,6 +282,7 @@ static const AVOption scale_vaapi_options[] = {
> { "decrease", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = 1 }, 0, 0, FLAGS, .unit = "force_oar" },
> { "increase", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = 2 }, 0, 0, FLAGS, .unit = "force_oar" },
> { "force_divisible_by", "enforce that the output resolution is divisible by a defined integer when force_original_aspect_ratio is used", OFFSET(force_divisible_by), AV_OPT_TYPE_INT, { .i64 = 1}, 1, 256, FLAGS },
> + { "reset_sar", "scale output frame to have square pixels", OFFSET(reset_sar), AV_OPT_TYPE_BOOL, { .i64 = 0}, 0, 1, FLAGS },
>
> { NULL },
> };
> diff --git a/libavfilter/vf_scale_vt.c b/libavfilter/vf_scale_vt.c
> index 162ea8fcec..c024717bc2 100644
> --- a/libavfilter/vf_scale_vt.c
> +++ b/libavfilter/vf_scale_vt.c
> @@ -248,7 +248,7 @@ static int scale_vt_config_output(AVFilterLink *outlink)
> if (err < 0)
> return err;
>
> - ff_scale_adjust_dimensions(inlink, &s->output_width, &s->output_height, 0, 1);
> + ff_scale_adjust_dimensions(inlink, &s->output_width, &s->output_height, 0, 1, 1.f);
>
> outlink->w = s->output_width;
> outlink->h = s->output_height;
> diff --git a/libavfilter/vf_scale_vulkan.c b/libavfilter/vf_scale_vulkan.c
> index 6f99769d81..1d6492e213 100644
> --- a/libavfilter/vf_scale_vulkan.c
> +++ b/libavfilter/vf_scale_vulkan.c
> @@ -336,7 +336,7 @@ static int scale_vulkan_config_output(AVFilterLink *outlink)
> if (err < 0)
> return err;
>
> - ff_scale_adjust_dimensions(inlink, &vkctx->output_width, &vkctx->output_height, 0, 1);
> + ff_scale_adjust_dimensions(inlink, &vkctx->output_width, &vkctx->output_height, 0, 1, 1.f);
>
> outlink->w = vkctx->output_width;
> outlink->h = vkctx->output_height;
More information about the ffmpeg-devel
mailing list