On Thu, 28 Nov 2013 15:03:04 +0100, Vittorio Giovara
<[email protected]> wrote:
> ---
> Changelog | 1 +
> doc/filters.texi | 19 ++
> libavfilter/Makefile | 1 +
> libavfilter/allfilters.c | 1 +
> libavfilter/version.h | 4 +-
> libavfilter/vf_framepack.c | 488
> ++++++++++++++++++++++++++++++++++++++++++++
> 6 files changed, 512 insertions(+), 2 deletions(-)
> create mode 100644 libavfilter/vf_framepack.c
>
> diff --git a/Changelog b/Changelog
> index 78b3338..5a47b18 100644
> --- a/Changelog
> +++ b/Changelog
> @@ -51,6 +51,7 @@ version 10:
> - support for decoding through VDPAU in avconv (the -hwaccel option)
> - remove mp3_header_(de)compress bitstream filters
> - codec level stereoscopic metadata handling
> +- framepack filter
>
>
> version 9:
> diff --git a/doc/filters.texi b/doc/filters.texi
> index b32aad1..04779bc 100644
> --- a/doc/filters.texi
> +++ b/doc/filters.texi
> @@ -1252,6 +1252,25 @@ frames with a negative PTS.
>
> @end table
>
> +@section framepack
> +
> +Packs two different video streams into a stereoscopic video. The two videos
> +should have the same size and framerate; you may conveniently adjust those
> +values with the @ref{scale} and @ref{fps} filters. If one video lasts longer
> +than the other, the last frame is repeated.
> +
> +This filter also sets proper signalling on supported codecs.
> +
> +This filter accepts the following named parameters:
> +@table @option
> +
> +@item format
> +Desired framepacking format. Supported values are @var{2d} (default),
> +@var{sbs}, @var{sbsqnx}, @var{tab}, @var{lines}, @var{columns},
> @var{frameseq}, @var{check}.
> +See a detailed format description in @file{libavutil/stereo3d.h}.
> +
> +@end table
> +
> @anchor{frei0r}
> @section frei0r
>
> diff --git a/libavfilter/Makefile b/libavfilter/Makefile
> index 96fa8c0..92c1561 100644
> --- a/libavfilter/Makefile
> +++ b/libavfilter/Makefile
> @@ -54,6 +54,7 @@ OBJS-$(CONFIG_FADE_FILTER) += vf_fade.o
> OBJS-$(CONFIG_FIELDORDER_FILTER) += vf_fieldorder.o
> OBJS-$(CONFIG_FORMAT_FILTER) += vf_format.o
> OBJS-$(CONFIG_FPS_FILTER) += vf_fps.o
> +OBJS-$(CONFIG_FRAMEPACK_FILTER) += vf_framepack.o
> OBJS-$(CONFIG_FREI0R_FILTER) += vf_frei0r.o
> OBJS-$(CONFIG_GRADFUN_FILTER) += vf_gradfun.o
> OBJS-$(CONFIG_HFLIP_FILTER) += vf_hflip.o
> diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
> index f041f5c..9702a0a 100644
> --- a/libavfilter/allfilters.c
> +++ b/libavfilter/allfilters.c
> @@ -74,6 +74,7 @@ void avfilter_register_all(void)
> REGISTER_FILTER(FIELDORDER, fieldorder, vf);
> REGISTER_FILTER(FORMAT, format, vf);
> REGISTER_FILTER(FPS, fps, vf);
> + REGISTER_FILTER(FRAMEPACK, framepack, vf);
> REGISTER_FILTER(FREI0R, frei0r, vf);
> REGISTER_FILTER(GRADFUN, gradfun, vf);
> REGISTER_FILTER(HFLIP, hflip, vf);
> diff --git a/libavfilter/version.h b/libavfilter/version.h
> index bc079b4..3f7d9dc 100644
> --- a/libavfilter/version.h
> +++ b/libavfilter/version.h
> @@ -30,8 +30,8 @@
> #include "libavutil/avutil.h"
>
> #define LIBAVFILTER_VERSION_MAJOR 3
> -#define LIBAVFILTER_VERSION_MINOR 11
> -#define LIBAVFILTER_VERSION_MICRO 1
> +#define LIBAVFILTER_VERSION_MINOR 12
> +#define LIBAVFILTER_VERSION_MICRO 0
>
> #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \
> LIBAVFILTER_VERSION_MINOR, \
> diff --git a/libavfilter/vf_framepack.c b/libavfilter/vf_framepack.c
> new file mode 100644
> index 0000000..35f5656
> --- /dev/null
> +++ b/libavfilter/vf_framepack.c
> @@ -0,0 +1,488 @@
> +/*
> + * Copyright (c) 2013 Vittorio Giovara
> + *
> + * This file is part of Libav.
> + *
> + * Libav is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU Lesser General Public
> + * License as published by the Free Software Foundation; either
> + * version 2.1 of the License, or (at your option) any later version.
> + *
> + * Libav is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + * Lesser General Public License for more details.
> + *
> + * You should have received a copy of the GNU Lesser General Public
> + * License along with Libav; if not, write to the Free Software
> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301
> USA
> + */
> +
> +/**
> + * @file
> + * Generate a frame packed video, by combining two views in a single surface.
> + */
> +
> +#include "libavutil/common.h"
> +#include "libavutil/opt.h"
> +#include "libavutil/pixdesc.h"
> +#include "libavutil/rational.h"
> +#include "libavutil/stereo3d.h"
> +
> +#include "avfilter.h"
> +#include "formats.h"
> +#include "internal.h"
> +#include "video.h"
> +
> +#define LEFT_VIEW 0
> +#define RIGHT_VIEW 1
> +
> +typedef struct FramepackContext {
> + const AVClass *class;
> +
> + const AVPixFmtDescriptor *pix_desc; ///< agreed pixel format
> +
> + enum AVStereo3DType format; ///< frame packed output
> +
> + AVFrame *left; ///< left input frame
> + AVFrame *right; ///< right input frame
You should be able to simplify the code by using an array[2] indexed with
LEFT/RIGHT_VIEW
> +} FramepackContext;
> +
> +static const enum AVPixelFormat formats_supported[] = {
> + AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P,
> + AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUVJ420P,
> + AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P,
> + AV_PIX_FMT_NONE
> +};
> +
> +static int query_formats(AVFilterContext *ctx)
> +{
> + ff_set_common_formats(ctx, ff_make_format_list(formats_supported));
> + return 0;
> +}
> +
> +static av_cold void framepack_uninit(AVFilterContext *ctx)
> +{
> + FramepackContext *s = ctx->priv;
> +
> + // clean any leftover frame
> + av_frame_free(&s->left);
> + av_frame_free(&s->right);
> +
> + av_opt_free(s);
Not needed, lavfi does that for you
> +}
> +
> +static int config_input_left(AVFilterLink *inlink)
> +{
> + FramepackContext *s = inlink->dst->priv;
> +
> + // save pixel format which has to match the other one
> + s->pix_desc = av_pix_fmt_desc_get(inlink->format);
> +
> + return 0;
> +}
> +
> +static int config_input_right(AVFilterLink *inlink)
> +{
> + FramepackContext *s = inlink->dst->priv;
> + const AVPixFmtDescriptor *right_desc =
> av_pix_fmt_desc_get(inlink->format);
> +
> + if(!s->pix_desc || !right_desc)
> + return AVERROR_BUG;
> +
> + // check input format
> + if (s->pix_desc != right_desc) {
> + av_log(inlink->dst, AV_LOG_ERROR,
> + "Left and right color spaces differ (%s vs %s).\n",
> + s->pix_desc->name, right_desc->name);
> + return AVERROR_INVALIDDATA;
> + }
The inputs should be treated symmetrically, there is no guarantee in which order
they are initialized. And the check is not needed anyway,
ff_set_common_formats() makes sure the formats are the same on all pads.
> +
> + return 0;
> +}
> +
> +static int config_output(AVFilterLink *outlink)
> +{
> + AVFilterContext *ctx = outlink->src;
> + FramepackContext *s = outlink->src->priv;
> +
> + int width = ctx->inputs[LEFT_VIEW]->w;
> + int height = ctx->inputs[LEFT_VIEW]->h;
> + AVRational time_base = ctx->inputs[LEFT_VIEW]->time_base;
> +
> + // check size and fps match on the other input
> + if (width != ctx->inputs[RIGHT_VIEW]->w ||
> + height != ctx->inputs[RIGHT_VIEW]->h) {
> + av_log(ctx, AV_LOG_ERROR,
> + "Left and right sizes differ (%dx%d vs %dx%d).\n",
> + width, height,
> + ctx->inputs[RIGHT_VIEW]->w, ctx->inputs[RIGHT_VIEW]->h);
> + return AVERROR_INVALIDDATA;
> + } else if (av_cmp_q(time_base, ctx->inputs[RIGHT_VIEW]->time_base) != 0)
> {
> + av_log(ctx, AV_LOG_ERROR,
> + "Left and right framerates differ (%d/%d vs %d/%d).\n",
> + time_base.num, time_base.den,
> + ctx->inputs[RIGHT_VIEW]->time_base.num,
> + ctx->inputs[RIGHT_VIEW]->time_base.den);
> + return AVERROR_INVALIDDATA;
> + }
> +
> + // modify output properties as needed
> + switch (s->format) {
> + case AV_STEREO3D_2D:
> + av_log(ctx, AV_LOG_WARNING,
> + "No frame packing mode selected, setting metadata only.\n");
> + break;
> + case AV_STEREO3D_FRAMESEQUENCE:
> + time_base.den *= 2;
> + break;
> + case AV_STEREO3D_COLUMNS:
> + case AV_STEREO3D_SIDEBYSIDE:
> + case AV_STEREO3D_SIDEBYSIDE_QUINCUNX:
> + width *= 2;
> + break;
> + case AV_STEREO3D_LINES:
> + case AV_STEREO3D_TOPBOTTOM:
> + height *= 2;
> + break;
> + case AV_STEREO3D_CHECKERBOARD:
> + // same w/h/tb
> + break;
> + }
> +
> + outlink->w = width;
> + outlink->h = height;
> + outlink->time_base = time_base;
> +
> + return 0;
> +}
> +
> +static void pack_sidebyside_frame(FramepackContext *s,
> + AVFrame *dst,
> + int interleaved)
> +{
> + int plane, i;
> + int length = s->left->width;
> + int lines = s->left->height;
> +
> + for (plane = 0; plane < s->pix_desc->nb_components; plane++) {
> + const uint8_t *leftp = s->left->data[plane];
> + const uint8_t *rightp = s->right->data[plane];
> + uint8_t *dstp = dst->data[plane];
> +
> + if (plane == 1 || plane == 2) {
> + length = -(-s->left->width >> s->pix_desc->log2_chroma_w);
> + lines = -(-s->left->height >> s->pix_desc->log2_chroma_h);
> + }
> +
> + if (!interleaved) {
> + for (i = 0; i < lines; i++) {
> + memcpy(dstp, leftp, length);
> + memcpy(dstp + length, rightp, length);
> +
> + dstp += dst->linesize[plane];
> + leftp += s->left->linesize[plane];
> + rightp += s->right->linesize[plane];
> + }
Can't av_image_copy() handle the non-interleaved case?
> + } else {
> + for (i = 0; i < lines; i++) {
> + int j;
> + int k = 0;
> +
> + for (j = 0; j < length; j++) {
> + dstp[k++] = leftp[j];
> + dstp[k++] = rightp[j];
> + }
> +
> + dstp += dst->linesize[plane];
> + leftp += s->left->linesize[plane];
> + rightp += s->right->linesize[plane];
> + }
> + }
> + }
> +
> + return;
> +}
> +
> +static void pack_topbottom_frame(FramepackContext *s,
> + AVFrame *dst,
> + int interleaved)
> +{
> + int plane, i;
> + int step = interleaved ? 2 : 1;
> + int length = s->left->width;
> + int lines = s->left->height;
> +
> + for (plane = 0; plane < s->pix_desc->nb_components; plane++) {
> + const uint8_t *srcp;
> + uint8_t *dstp;
> +
> + if (plane == 1 || plane == 2) {
> + length = -(-s->left->width >> s->pix_desc->log2_chroma_w);
> + lines = -(-s->left->height >> s->pix_desc->log2_chroma_h);
> + }
> +
> + // copy first frame line by line
> + srcp = s->left->data[plane];
> + dstp = dst->data[plane];
> + for (i = 0; i < lines; i ++) {
> + memcpy(dstp, srcp, length);
> + dstp += dst->linesize[plane] * step;
> + srcp += s->left->linesize[plane];
> + }
> +
> + // copy second frame line by line
> + srcp = s->right->data[plane];
> + if (interleaved) {
> + // reset destination pointer and offset one line
> + dstp = dst->data[plane] + dst->linesize[plane];
> + }
> + for (i = 0; i < lines; i ++) {
> + memcpy(dstp, srcp, length);
> + dstp += dst->linesize[plane] * step;
> + srcp += s->right->linesize[plane];
> + }
> + }
Can't this be replaced by av_image_copy()?
> +
> + return;
> +}
> +
> +static void pack_checkerboard_frame(FramepackContext *s, AVFrame *dst)
> +{
> + int plane, i;
> + int length = s->left->width;
> + int lines = s->left->height;
> +
> + for (plane = 0; plane < s->pix_desc->nb_components; plane++) {
> + const uint8_t *leftp = s->left->data[plane];
> + const uint8_t *rightp = s->right->data[plane];
> + uint8_t *dstp = dst->data[plane];
> +
> + if (plane == 1 || plane == 2) {
> + length = -(-s->left->width >> s->pix_desc->log2_chroma_w);
> + lines = -(-s->left->height >> s->pix_desc->log2_chroma_h);
> + }
> +
> + // alternatively copy one pixel, changing source at each line
> + for (i = 0; i < lines; i++) {
> + int k;
> + for (k = 0; k < length; k += 2) {
> + dstp[k] = i % 2 ? rightp[k] : leftp[k];
> + dstp[k + 1] = i % 2 ? leftp[k + 1] : rightp[k + 1];
> + }
This code looks shady. Should k not run over input length * 2? Or does this
packing lose half the resolution?
> +
> + dstp += dst->linesize[plane];
> + leftp += s->left->linesize[plane];
> + rightp += s->right->linesize[plane];
> + }
> + }
> +
> + return;
> +}
> +
> +static int set_side_data(AVFrame *out, enum AVStereo3DType format)
> +{
> + AVStereo3D *stereo = av_stereo3d_create_side_data(out);
> + if (!stereo)
> + return AVERROR(ENOMEM);
> +
> + stereo->type = format;
> +
> + return 0;
> +}
> +
> +
> +static av_always_inline void pack_frame(FramepackContext *s, AVFrame *dst)
> +{
> + switch (s->format) {
> + case AV_STEREO3D_COLUMNS:
> + pack_sidebyside_frame(s, dst, 1);
> + break;
> + case AV_STEREO3D_SIDEBYSIDE:
> + case AV_STEREO3D_SIDEBYSIDE_QUINCUNX:
> + pack_sidebyside_frame(s, dst, 0);
> + break;
> + case AV_STEREO3D_LINES:
> + pack_topbottom_frame(s, dst, 1);
> + break;
> + case AV_STEREO3D_TOPBOTTOM:
> + pack_topbottom_frame(s, dst, 0);
> + break;
> + case AV_STEREO3D_CHECKERBOARD:
> + pack_checkerboard_frame(s, dst);
> + break;
> + }
> +
> + return;
> +}
> +
> +static int filter_frame_left(AVFilterLink *inlink, AVFrame *frame)
> +{
> + FramepackContext *s = inlink->dst->priv;
> +
> + s->left = frame;
> +
> + return 0;
> +}
> +
> +static int filter_frame_right(AVFilterLink *inlink, AVFrame *frame)
> +{
> + FramepackContext *s = inlink->dst->priv;
> +
> + s->right = frame;
> +
> + return 0;
> +}
> +
> +static int request_frame(AVFilterLink *outlink)
> +{
> + AVFilterContext *ctx = outlink->src;
> + FramepackContext *s = ctx->priv;
> + int ret, retl, retr;
> + AVFrame *out;
> + static uint64_t double_pts;
static? really? =p
> +
> + /* get a frame on the left input, on EOF reuse previous */
> + if (!s->left) {
> + retl = ff_request_frame(ctx->inputs[LEFT_VIEW]);
> + if (retl < 0 && retl != AVERROR_EOF)
> + return retl;
> + }
> + /* get a new frame on the right input, on EOF reuse previous */
> + if (!s->right) {
> + retr = ff_request_frame(ctx->inputs[RIGHT_VIEW]);
> + if (retr < 0 && retr != AVERROR_EOF)
> + return retr;
> + }
> + /* when both return values are EOF, it means that both
> + * videos are fully processed and we can stop */
> + if (retl == retr && retl == AVERROR_EOF)
> + return retl;
> +
> + if (s->format == AV_STEREO3D_FRAMESEQUENCE) {
> + AVFrame *cur;
> + int i;
> + double_pts = s->left->pts;
> +
> + for (i = 0; i < 2; i++) {
> + cur = i == 0 ? s->left : s->right;
> + out = av_frame_clone(cur);
> + if (!out)
> + return AVERROR(ENOMEM);
> +
> + out->pts = double_pts++;
> + ret = set_side_data(out, s->format);
> + if (ret < 0)
> + return ret;
> +
> + ret = ff_filter_frame(outlink, out);
> + if (ret < 0)
> + return ret;
> + }
> + if (retl != AVERROR_EOF)
> + av_frame_free(&s->left);
> + if (retr != AVERROR_EOF)
> + av_frame_free(&s->right);
> + return ret;
> + } else if (s->format == AV_STEREO3D_2D) {
> + out = av_frame_clone(s->left);
How is that supposed to work?
The caller connects two streams, but only the left one is actually read?
Looks awkward/hacky to me, perhaps it'd be better to have a separate filter.
> + if (!out)
> + return AVERROR(ENOMEM);
> + } else {
> + out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
> + if (!out)
> + return AVERROR(ENOMEM);
> +
> + pack_frame(s, out);
> + }
> +
> + ret = av_frame_copy_props(out, s->left);
> + if (ret < 0)
> + return ret;
> +
> + ret = set_side_data(out, s->format);
> + if (ret < 0)
> + return ret;
> +
> + ret = ff_filter_frame(outlink, out);
> + if (ret < 0)
> + return ret;
> +
> + // cleanup keeping a valid frame on EOF
> + if (retl != AVERROR_EOF)
> + av_frame_free(&s->left);
> + if (retr != AVERROR_EOF)
> + av_frame_free(&s->right);
You could move the cleanup before ff_filter_frame, then just return
ff_filter_frame()
--
Anton Khirnov
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
