X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavfilter%2Fvf_select.c;h=b187fa3d7cd5425fcd91cbae4de63eaac8ba7fdb;hb=9c1d649ce57c2b707b5a2eab47f05dd1833bda22;hp=2ffa49678816a9e2baa5ae89f244fb50c31c3faa;hpb=9eb99ab25b5ec9af9e575371c9ef84a37c63b3e5;p=ffmpeg diff --git a/libavfilter/vf_select.c b/libavfilter/vf_select.c index 2ffa4967881..b187fa3d7cd 100644 --- a/libavfilter/vf_select.c +++ b/libavfilter/vf_select.c @@ -25,7 +25,11 @@ #include "libavutil/eval.h" #include "libavutil/fifo.h" +#include "libavcodec/dsputil.h" #include "avfilter.h" +#include "formats.h" +#include "internal.h" +#include "video.h" static const char *const var_names[] = { "TB", ///< timebase @@ -61,6 +65,8 @@ static const char *const var_names[] = { "key", ///< tell if the frame is a key frame "pos", ///< original position in the file of the frame + "scene", + NULL }; @@ -98,6 +104,8 @@ enum var_name { VAR_KEY, VAR_POS, + VAR_SCENE, + VAR_VARS_NB }; @@ -106,12 +114,17 @@ enum var_name { typedef struct { AVExpr *expr; double var_values[VAR_VARS_NB]; + int do_scene_detect; ///< 1 if the expression requires scene detection variables, 0 otherwise + AVCodecContext *avctx; ///< codec context required for the DSPContext (scene detect only) + DSPContext c; ///< context providing optimized SAD methods (scene detect only) + double prev_mafd; ///< previous MAFD (scene detect only) + AVFilterBufferRef *prev_picref; ///< previous frame (scene detect only) double select; int cache_frames; AVFifoBuffer *pending_frames; ///< FIFO buffer of video frames } SelectContext; -static av_cold int init(AVFilterContext *ctx, const char *args, void *opaque) +static av_cold int init(AVFilterContext *ctx, const char *args) { SelectContext *select = ctx->priv; int ret; @@ -127,6 +140,8 @@ static av_cold int init(AVFilterContext *ctx, const char *args, void *opaque) av_log(ctx, AV_LOG_ERROR, "Failed to allocate pending frames buffer.\n"); return AVERROR(ENOMEM); } + + select->do_scene_detect = args && strstr(args, "scene"); return 0; } @@ -159,9 +174,49 @@ static int config_input(AVFilterLink *inlink) select->var_values[VAR_INTERLACE_TYPE_T] = INTERLACE_TYPE_T; select->var_values[VAR_INTERLACE_TYPE_B] = INTERLACE_TYPE_B; + if (select->do_scene_detect) { + select->avctx = avcodec_alloc_context3(NULL); + if (!select->avctx) + return AVERROR(ENOMEM); + dsputil_init(&select->c, select->avctx); + } return 0; } +static double get_scene_score(AVFilterContext *ctx, AVFilterBufferRef *picref) +{ + double ret = 0; + SelectContext *select = ctx->priv; + AVFilterBufferRef *prev_picref = select->prev_picref; + + if (prev_picref && + picref->video->h == prev_picref->video->h && + picref->video->w == prev_picref->video->w && + picref->linesize[0] == prev_picref->linesize[0]) { + int x, y; + int64_t sad; + double mafd, diff; + uint8_t *p1 = picref->data[0]; + uint8_t *p2 = prev_picref->data[0]; + const int linesize = picref->linesize[0]; + + for (sad = y = 0; y < picref->video->h; y += 8) + for (x = 0; x < linesize; x += 8) + sad += select->c.sad[1](select, + p1 + y * linesize + x, + p2 + y * linesize + x, + linesize, 8); + emms_c(); + mafd = sad / (picref->video->h * picref->video->w * 3); + diff = fabs(mafd - select->prev_mafd); + ret = av_clipf(FFMIN(mafd, diff) / 100., 0, 1); + select->prev_mafd = mafd; + avfilter_unref_buffer(prev_picref); + } + select->prev_picref = avfilter_ref_buffer(picref, ~0); + return ret; +} + #define D2TS(d) (isnan(d) ? AV_NOPTS_VALUE : (int64_t)(d)) #define TS2D(ts) ((ts) == AV_NOPTS_VALUE ? NAN : (double)(ts)) @@ -171,6 +226,8 @@ static int select_frame(AVFilterContext *ctx, AVFilterBufferRef *picref) AVFilterLink *inlink = ctx->inputs[0]; double res; + if (select->do_scene_detect) + select->var_values[VAR_SCENE] = get_scene_score(ctx, picref); if (isnan(select->var_values[VAR_START_PTS])) select->var_values[VAR_START_PTS] = TS2D(picref->pts); if (isnan(select->var_values[VAR_START_T])) @@ -228,7 +285,7 @@ static void start_frame(AVFilterLink *inlink, AVFilterBufferRef *picref) sizeof(picref), NULL); return; } - avfilter_start_frame(inlink->dst->outputs[0], avfilter_ref_buffer(picref, ~0)); + ff_start_frame(inlink->dst->outputs[0], avfilter_ref_buffer(picref, ~0)); } } @@ -237,20 +294,18 @@ static void draw_slice(AVFilterLink *inlink, int y, int h, int slice_dir) SelectContext *select = inlink->dst->priv; if (select->select && !select->cache_frames) - avfilter_draw_slice(inlink->dst->outputs[0], y, h, slice_dir); + ff_draw_slice(inlink->dst->outputs[0], y, h, slice_dir); } static void end_frame(AVFilterLink *inlink) { SelectContext *select = inlink->dst->priv; - AVFilterBufferRef *picref = inlink->cur_buf; if (select->select) { if (select->cache_frames) return; - avfilter_end_frame(inlink->dst->outputs[0]); + ff_end_frame(inlink->dst->outputs[0]); } - avfilter_unref_buffer(picref); } static int request_frame(AVFilterLink *outlink) @@ -263,15 +318,15 @@ static int request_frame(AVFilterLink *outlink) if (av_fifo_size(select->pending_frames)) { AVFilterBufferRef *picref; av_fifo_generic_read(select->pending_frames, &picref, sizeof(picref), NULL); - avfilter_start_frame(outlink, avfilter_ref_buffer(picref, ~0)); - avfilter_draw_slice(outlink, 0, outlink->h, 1); - avfilter_end_frame(outlink); + ff_start_frame(outlink, avfilter_ref_buffer(picref, ~0)); + ff_draw_slice(outlink, 0, outlink->h, 1); + ff_end_frame(outlink); avfilter_unref_buffer(picref); return 0; } while (!select->select) { - int ret = avfilter_request_frame(inlink); + int ret = ff_request_frame(inlink); if (ret < 0) return ret; } @@ -286,12 +341,12 @@ static int poll_frame(AVFilterLink *outlink) int count, ret; if (!av_fifo_size(select->pending_frames)) { - if ((count = avfilter_poll_frame(inlink)) <= 0) + if ((count = ff_poll_frame(inlink)) <= 0) return count; /* request frame from input, and apply select condition to it */ select->cache_frames = 1; while (count-- && av_fifo_space(select->pending_frames)) { - ret = avfilter_request_frame(inlink); + ret = ff_request_frame(inlink); if (ret < 0) break; } @@ -314,6 +369,28 @@ static av_cold void uninit(AVFilterContext *ctx) avfilter_unref_buffer(picref); av_fifo_free(select->pending_frames); select->pending_frames = NULL; + + if (select->do_scene_detect) { + avfilter_unref_bufferp(&select->prev_picref); + avcodec_close(select->avctx); + av_freep(&select->avctx); + } +} + +static int query_formats(AVFilterContext *ctx) +{ + SelectContext *select = ctx->priv; + + if (!select->do_scene_detect) { + return ff_default_query_formats(ctx); + } else { + static const enum PixelFormat pix_fmts[] = { + PIX_FMT_RGB24, PIX_FMT_BGR24, + PIX_FMT_NONE + }; + ff_set_common_formats(ctx, ff_make_format_list(pix_fmts)); + } + return 0; } AVFilter avfilter_vf_select = { @@ -321,12 +398,13 @@ AVFilter avfilter_vf_select = { .description = NULL_IF_CONFIG_SMALL("Select frames to pass in output."), .init = init, .uninit = uninit, + .query_formats = query_formats, .priv_size = sizeof(SelectContext), .inputs = (const AVFilterPad[]) {{ .name = "default", .type = AVMEDIA_TYPE_VIDEO, - .get_video_buffer = avfilter_null_get_video_buffer, + .get_video_buffer = ff_null_get_video_buffer, .config_props = config_input, .start_frame = start_frame, .draw_slice = draw_slice,