[FFmpeg-devel] [PATCH] avfilter: add ANSNR filter
Betty Wu
lumosomul at gmail.com
Mon Apr 3 18:00:42 EEST 2017
A new filter ANSNR is added. libavfilter/Makefile is changed.
Run 'ffmpeg -i input1 -i input2 -lavfi ansnr -f null -' to get an overall score while per-frame value is stored but not printed.
This implementation is for constructing the vmaf filter later since ANSNR is one of individual tools used in vmaf.
Signed-off-by: Betty Wu <lumosomul at gmail.com>
---
libavfilter/Makefile | 1 +
libavfilter/allfilters.c | 1 +
libavfilter/vf_ansnr.c | 425 +++++++++++++++++++++++++++++++++++++++++++++++
3 files changed, 427 insertions(+)
create mode 100644 libavfilter/vf_ansnr.c
diff --git a/libavfilter/Makefile b/libavfilter/Makefile
index 9c15ed62d2..5416e0f34f 100644
--- a/libavfilter/Makefile
+++ b/libavfilter/Makefile
@@ -123,6 +123,7 @@ OBJS-$(CONFIG_ANULLSINK_FILTER) += asink_anullsink.o
# video filters
OBJS-$(CONFIG_ALPHAEXTRACT_FILTER) += vf_extractplanes.o
OBJS-$(CONFIG_ALPHAMERGE_FILTER) += vf_alphamerge.o
+OBJS-$(CONFIG_ANSNR_FILTER) += vf_ansnr.o dualinput.o framesync.o
OBJS-$(CONFIG_ASS_FILTER) += vf_subtitles.o
OBJS-$(CONFIG_ATADENOISE_FILTER) += vf_atadenoise.o
OBJS-$(CONFIG_AVGBLUR_FILTER) += vf_avgblur.o
diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
index 64b634e8f3..fbd9ec026e 100644
--- a/libavfilter/allfilters.c
+++ b/libavfilter/allfilters.c
@@ -134,6 +134,7 @@ static void register_all(void)
REGISTER_FILTER(ALPHAEXTRACT, alphaextract, vf);
REGISTER_FILTER(ALPHAMERGE, alphamerge, vf);
+ REGISTER_FILTER(ANSNR, ansnr, vf);
REGISTER_FILTER(ASS, ass, vf);
REGISTER_FILTER(ATADENOISE, atadenoise, vf);
REGISTER_FILTER(AVGBLUR, avgblur, vf);
diff --git a/libavfilter/vf_ansnr.c b/libavfilter/vf_ansnr.c
new file mode 100644
index 0000000000..cfea6efbd4
--- /dev/null
+++ b/libavfilter/vf_ansnr.c
@@ -0,0 +1,425 @@
+/*
+ * Copyright (c) 2017 Betty Wu
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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.
+ *
+ * FFmpeg 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 FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/*
+ * @file
+ * Caculate the ANSNR between two input videos.
+ * @author Betty Wu
+ */
+
+#include "libavutil/avstring.h"
+#include "libavutil/opt.h"
+#include "libavutil/pixdesc.h"
+#include "avfilter.h"
+#include "dualinput.h"
+#include "drawutils.h"
+#include "formats.h"
+#include "internal.h"
+#include "video.h"
+
+#define OPT_RANGE_PIXEL_OFFSET -128
+
+typedef double number_t;
+
+typedef struct ANSNRContext {
+ const AVClass *class;
+ FFDualInputContext dinput;
+ uint64_t nb_frames;
+ FILE *stats_file;
+ char *stats_file_str;
+ int stats_version;
+ int stats_header_written;
+ int stats_add_max;
+ int is_rgb;
+ uint8_t rgba_map[4];
+ double score;
+ double score_total;
+ char comps[4];
+ int nb_components;
+ int planewidth[4];
+ int planeheight[4];
+
+} ANSNRContext;
+#define OFFSET(x) offsetof(ANSNRContext, x)
+#define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
+
+static const AVOption ansnr_options[] = {
+ {"stats_file", "Set file where to store per-frame difference information", OFFSET(stats_file_str), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, FLAGS },
+ {"f", "Set file where to store per-frame difference information", OFFSET(stats_file_str), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, FLAGS },
+ { NULL }
+};
+AVFILTER_DEFINE_CLASS(ansnr);
+
+const int ans_filter3_stride = 3;
+const int ans_filter5_stride = 5;
+
+const double ans_filter3[ans_filter3_stride*ans_filter3_stride] = {
+ 1.0 / 16.0, 2.0 / 16.0, 1.0 / 16.0,
+ 2.0 / 16.0, 4.0 / 16.0, 2.0 / 16.0,
+ 1.0 / 16.0, 2.0 / 16.0, 1.0 / 16.0
+};
+
+const double ans_filter5[ans_filter5_stride*ans_filter5_stride] = {
+ 2.0 / 571.0, 7.0 / 571.0, 12.0 / 571.0, 7.0 / 571.0, 2.0 / 571.0,
+ 7.0 / 571.0, 31.0 / 571.0, 52.0 / 571.0, 31.0 / 571.0, 7.0 / 571.0,
+ 12.0 / 571.0, 52.0 / 571.0, 127.0 / 571.0, 52.0 / 571.0, 12.0 / 571.0,
+ 7.0 / 571.0, 31.0 / 571.0, 52.0 / 571.0, 31.0 / 571.0, 7.0 / 571.0,
+ 2.0 / 571.0, 7.0 / 571.0, 12.0 / 571.0, 7.0 / 571.0, 2.0 / 571.0
+};
+
+static inline number_t pow_2(number_t base)
+{
+ return base*base;
+}
+
+static inline double get_ansnr(double score)
+{
+ return 10.0 * log10(score);
+}
+
+static inline double get_pow(double score)
+{
+ return pow(10, score/10.0);
+}
+
+static void ansnr_mse(const number_t *ref, const number_t *dis, number_t *img_sig, number_t *img_noise, int row, int col)
+{
+ int ref_stride = col;
+ int dis_stride = col;
+
+ *img_sig = 0;
+ *img_noise = 0;
+
+ for (int i = 0; i < row; ++i) {
+ for (int j = 0; j < col; ++j) {
+ int ref_index = i * ref_stride + j;
+ int dis_index = i * dis_stride + j;
+
+ *img_sig += pow_2(ref[ref_index]);
+ *img_noise += pow_2(ref[ref_index] - dis[dis_index]);
+ }
+ }
+}
+
+static int ansnr_filter(const uint8_t* src_image, number_t* dst_image, int img_row, int img_col, const double *filter, int stride)
+{
+ int filter_row = stride;
+ int filter_col = stride;
+ int start_row = filter_row / 2;
+ int start_col = filter_col / 2;
+ int imme_row = img_row + 2 * start_row;
+ int imme_col = img_col + 2 * start_col;
+
+ number_t **imme_image;
+ if(!(imme_image = (number_t **)av_malloc((size_t)imme_row * sizeof(number_t *))))
+ return AVERROR(ENOMEM);
+
+ for(int i = 0; i < imme_row; i++) {
+ if(!(imme_image[i] = (number_t *)av_malloc((size_t)imme_col * sizeof(number_t))))
+ return AVERROR(ENOMEM);
+ }
+
+ for (int i = 0; i < imme_row; i++) {
+ for (int j = 0; j < imme_col; j++) {
+ int src_i = i - start_row;
+ int src_j = j - start_col;
+
+ src_i = abs(src_i) >= img_row? 2 * img_row - abs(src_i) - 1 : abs(src_i);
+ src_j = abs(src_j) >= img_col? 2 * img_col - abs(src_j) - 1 : abs(src_j);
+
+ imme_image[i][j] = (number_t)src_image[src_i * img_col + src_j] + OPT_RANGE_PIXEL_OFFSET;
+ }
+ }
+
+ int end_row = imme_row -1 - start_row ;
+ int end_col = imme_col -1 - start_col ;
+
+ for (int i = start_row; i <= end_row; i++) {
+ for (int j = start_col; j <= end_col; j++) {
+ int dst_i = i - start_row ;
+ int dst_j = j - start_col ;
+ int dst_index = dst_i * img_col + dst_j;
+ dst_image[dst_index] = 0;
+
+ for (int m = 0; m < filter_row; m++) {
+ for (int n = 0; n < filter_col; n++) {
+ dst_image[dst_index] += (imme_image[dst_i + m][dst_j + n] * filter[m * filter_col + n]);
+ }
+ }
+ }
+ }
+
+ for(int i = 0; i < imme_row; i++)
+ av_free(imme_image[i]);
+ av_free(imme_image);
+
+ return 0;
+}
+
+static inline
+int compute_ansnr(const uint8_t *anc, const uint8_t *dis, int col, int row, double psnr_max, double *score, double *out)
+{
+ *score = 0.0;
+ *out = 0.0;
+
+ number_t *anc_after;
+ if(!(anc_after = (number_t *)av_malloc((size_t)row * col * sizeof(number_t)))) {
+ av_free(anc_after);
+ return AVERROR(ENOMEM);
+ }
+
+ number_t *dis_after;
+ if(!(dis_after = (number_t *)av_malloc((size_t)row * col * sizeof(number_t)))) {
+ av_free(dis_after);
+ return AVERROR(ENOMEM);
+ }
+
+ int ret;
+ ret = ansnr_filter(anc, anc_after, row, col, ans_filter3, ans_filter3_stride);
+ if(ret < 0)
+ return ret;
+ ret = ansnr_filter(dis, dis_after, row, col, ans_filter5, ans_filter5_stride);
+ if(ret < 0)
+ return ret;
+
+ number_t sig, noise;
+
+ ansnr_mse(anc_after, dis_after, &sig, &noise, row, col);
+ *score = noise==0 ? psnr_max : get_ansnr((double)sig / noise);
+ *out = get_pow(*score);
+
+ av_free(anc_after);
+ av_free(dis_after);
+
+ return 0;
+}
+
+static void set_meta(AVDictionary **metadata, const char *key, char comp, float d)
+{
+ char value[128];
+ snprintf(value, sizeof(value), "%0.2f", d);
+ if (comp) {
+ char key2[128];
+ snprintf(key2, sizeof(key2), "%s%c", key, comp);
+ av_dict_set(metadata, key2, value, 0);
+ } else {
+ av_dict_set(metadata, key, value, 0);
+ }
+}
+
+static int query_formats(AVFilterContext *ctx)
+{
+ static const enum AVPixelFormat pix_fmts[] = {
+ AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY16,
+#define PF_NOALPHA(suf) AV_PIX_FMT_YUV420##suf, AV_PIX_FMT_YUV422##suf, AV_PIX_FMT_YUV444##suf
+#define PF_ALPHA(suf) AV_PIX_FMT_YUVA420##suf, AV_PIX_FMT_YUVA422##suf, AV_PIX_FMT_YUVA444##suf
+#define PF(suf) PF_NOALPHA(suf), PF_ALPHA(suf)
+ PF(P), PF(P9), PF(P10), PF_NOALPHA(P12), PF_NOALPHA(P14), PF(P16),
+ AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
+ AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P,
+ AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_YUVJ444P,
+ AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
+ AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
+ AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP16,
+ AV_PIX_FMT_NONE
+ };
+
+ AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
+ if (!fmts_list)
+ return AVERROR(ENOMEM);
+ return ff_set_common_formats(ctx, fmts_list);
+}
+
+static AVFrame *do_ansnr(AVFilterContext *ctx, AVFrame *main,
+ const AVFrame *ref)
+{
+ ANSNRContext *s = ctx->priv;
+ AVDictionary **metadata = avpriv_frame_get_metadatap(main);
+ double psnr_max;
+ char *frame_format = (char *)av_get_pix_fmt_name(main->format);
+
+ if( !strcmp(frame_format,"yuv420p") || !strcmp(frame_format,"yuv422p") || !strcmp(frame_format,"yuv444p"))
+ psnr_max = 60;
+ if( !strcmp(frame_format,"yuv420p10le") || !strcmp(frame_format,"yuv422p10le") || !strcmp(frame_format,"yuv444p10le"))
+ psnr_max = 72;
+
+ double score = 0.0;
+ double out = 0.0;
+
+ compute_ansnr(ref->data[0], main->data[0], s->planewidth[0], s->planeheight[0], psnr_max, &score, &out);
+ s->nb_frames++;
+ s->score = score;
+ s->score_total += out;
+
+ set_meta(metadata, "lavfi.ansnr.All", 0, score);
+
+ if (s->stats_file) {
+ fprintf(s->stats_file, "n:%"PRId64" ", s->nb_frames);
+ fprintf(s->stats_file, "All:%f\n", score);
+ }
+
+ return main;
+}
+
+static av_cold int init(AVFilterContext *ctx)
+{
+ ANSNRContext *s = ctx->priv;
+ s->score_total = 0;
+ s->score = 0;
+
+ if (s->stats_file_str) {
+ if (s->stats_version < 2 && s->stats_add_max) {
+ av_log(ctx, AV_LOG_ERROR,
+ "stats_add_max was specified but stats_version < 2.\n" );
+ return AVERROR(EINVAL);
+ }
+
+ if (!strcmp(s->stats_file_str, "-")) {
+ s->stats_file = stdout;
+ } else {
+ s->stats_file = fopen(s->stats_file_str, "w");
+ if (!s->stats_file) {
+ int err = AVERROR(errno);
+ char buf[128];
+ av_strerror(err, buf, sizeof(buf));
+ av_log(ctx, AV_LOG_ERROR, "Could not open stats file %s: %s\n",
+ s->stats_file_str, buf);
+ return err;
+ }
+ }
+ }
+
+ s->dinput.process = do_ansnr;
+ return 0;
+}
+
+static int config_input_ref(AVFilterLink *inlink)
+{
+ const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
+ AVFilterContext *ctx = inlink->dst;
+ ANSNRContext *s = ctx->priv;
+
+ s->nb_components = desc->nb_components;
+
+ if (ctx->inputs[0]->w != ctx->inputs[1]->w ||
+ ctx->inputs[0]->h != ctx->inputs[1]->h) {
+ av_log(ctx, AV_LOG_ERROR, "Width and height of input videos must be same.\n");
+ return AVERROR(EINVAL);
+ }
+ if (ctx->inputs[0]->format != ctx->inputs[1]->format) {
+ av_log(ctx, AV_LOG_ERROR, "Inputs must be of same pixel format.\n");
+ return AVERROR(EINVAL);
+ }
+ s->is_rgb = ff_fill_rgba_map(s->rgba_map, inlink->format) >= 0;
+ s->comps[0] = s->is_rgb ? 'r' : 'y' ;
+ s->comps[1] = s->is_rgb ? 'g' : 'u' ;
+ s->comps[2] = s->is_rgb ? 'b' : 'v' ;
+ s->comps[3] = 'a';
+
+ s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
+ s->planeheight[0] = s->planeheight[3] = inlink->h;
+ s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
+ s->planewidth[0] = s->planewidth[3] = inlink->w;
+
+ return 0;
+}
+
+static int config_output(AVFilterLink *outlink)
+{
+ AVFilterContext *ctx = outlink->src;
+ ANSNRContext *s = ctx->priv;
+ AVFilterLink *mainlink = ctx->inputs[0];
+ int ret;
+
+ outlink->w = mainlink->w;
+ outlink->h = mainlink->h;
+ outlink->time_base = mainlink->time_base;
+ outlink->sample_aspect_ratio = mainlink->sample_aspect_ratio;
+ outlink->frame_rate = mainlink->frame_rate;
+
+ if ((ret = ff_dualinput_init(ctx, &s->dinput)) < 0)
+ return ret;
+
+ return 0;
+}
+
+static int filter_frame(AVFilterLink *inlink, AVFrame *inpicref)
+{
+ ANSNRContext *s = inlink->dst->priv;
+ return ff_dualinput_filter_frame(&s->dinput, inlink, inpicref);
+}
+
+static int request_frame(AVFilterLink *outlink)
+{
+ ANSNRContext *s = outlink->src->priv;
+ return ff_dualinput_request_frame(&s->dinput, outlink);
+}
+
+static av_cold void uninit(AVFilterContext *ctx)
+{
+ ANSNRContext *s = ctx->priv;
+
+ if (s->nb_frames > 0) {
+ char buf[256];
+ buf[0] = 0;
+ av_log(ctx, AV_LOG_INFO, "ANSNR%s All:%f\n", buf, get_ansnr(s->score_total));
+ }
+
+ ff_dualinput_uninit(&s->dinput);
+
+ if (s->stats_file && s->stats_file != stdout)
+ fclose(s->stats_file);
+}
+
+static const AVFilterPad ansnr_inputs[] = {
+ {
+ .name = "main",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .filter_frame = filter_frame,
+ },{
+ .name = "reference",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .filter_frame = filter_frame,
+ .config_props = config_input_ref,
+ },
+ { NULL }
+};
+
+static const AVFilterPad ansnr_outputs[] = {
+ {
+ .name = "default",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .config_props = config_output,
+ .request_frame = request_frame,
+ },
+ { NULL }
+};
+
+AVFilter ff_vf_ansnr = {
+ .name = "ansnr",
+ .description = NULL_IF_CONFIG_SMALL("Calculate the ANSNR between two video streams."),
+ .init = init,
+ .uninit = uninit,
+ .query_formats = query_formats,
+ .priv_size = sizeof(ANSNRContext),
+ .priv_class = &ansnr_class,
+ .inputs = ansnr_inputs,
+ .outputs = ansnr_outputs,
+};
--
2.12.1
More information about the ffmpeg-devel
mailing list