[FFmpeg-devel] [PATCH] libavfilter: created a new filter that obtains the average peak signal-to-noise ratio (PSNR) of two input video files in YUV format.

Stefano Sabatini stefano.sabatini-lala at poste.it
Mon Jun 6 11:32:26 CEST 2011


On date Sunday 2011-06-05 21:00:35 +0200, Roger Pau Monn� encoded:
> 
> Signed-off-by: Roger Pau Monn? <roger.pau at entel.upc.edu>
> ---
>  libavfilter/vf_psnr.c |  272 +++++++++++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 272 insertions(+), 0 deletions(-)
>  create mode 100644 libavfilter/vf_psnr.c
> 
> diff --git a/libavfilter/vf_psnr.c b/libavfilter/vf_psnr.c
> new file mode 100644
> index 0000000..c6a0b78
> --- /dev/null
> +++ b/libavfilter/vf_psnr.c
> @@ -0,0 +1,272 @@
> +/*
> + * Copyright (c) 2011 Roger Pau Monné <roger.pau at entel.upc.edu>
> + *
> + * 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 PSNR between two input videos
> + * Based on the overlay filter
> + */
> +
> +#include "libavutil/pixdesc.h"
> +#include "avfilter.h"
> +
> +#undef fprintf
> +
> +typedef struct {
> +    AVFilterBufferRef *picref;
> +    double mse, min_mse, max_mse;
> +    int nb_frames;
> +    FILE *vstats_file;
> +    uint16_t *line1, *line2;
> +} PSNRContext;
> +
> +static inline int pow2(int base)
> +{
> +    return base*base;
> +}
> +
> +static inline double psnr(double mse, int nb_frames, int max)
> +{
> +    return 10.0*log((pow2(max))/(mse/nb_frames))/log(10.0);
> +}
> +
> +static inline void av_compute_images_mse(const uint8_t *ref_data[4], const uint8_t *data[4], const int linesizes[4], 
> +                                         int w, int h, const AVPixFmtDescriptor *desc, double mse[4], uint16_t *line1, 
> +                                         uint16_t *line2)

av_ is not required

> +{
> +    int i, c, j = w;
> +
> +    memset(mse, 0, sizeof(*mse)*4);
> +
> +    for (c = 0; c < desc->nb_components; c++) {
> +        int w1 = c == 1 || c == 2 ? w>>desc->log2_chroma_w : w;
> +        int h1 = c == 1 || c == 2 ? h>>desc->log2_chroma_h : h;
> +
> +        for (i = 0; i < h1; i++) {
> +            av_read_image_line(line1,
> +                                ref_data,
> +                                linesizes,
> +                                desc,
> +                                0, i, c, w1, 0);
> +            av_read_image_line(line2,
> +                                data,
> +                                linesizes,
> +                                desc,
> +                                0, i, c, w1, 0);
> +            for(j = 0; j < w1; j++)
> +                mse[c] += pow2(line1[j] - line2[j]);
> +        }
> +        mse[c] /= w1*h1;
> +    }
> +}
> +
> +
> +static av_cold int init(AVFilterContext *ctx, const char *args, void *opaque)
> +{
> +    PSNRContext *psnr_context = ctx->priv;

You can only use "psnr" here and below and save some chars, as it is
done in the other filters (and move psnr() -> get_psnr() or
compute_psnr()).

> +
> +    psnr_context->mse = psnr_context->nb_frames = 0;
> +    psnr_context->min_mse = psnr_context->max_mse = -1.0;
> +    psnr_context->picref = NULL;
> +
> +    if (args != NULL && strlen(args) > 0) {
> +        psnr_context->vstats_file = fopen(args, "w");
> +        if (!psnr_context->vstats_file) {
> +            av_log(ctx, AV_LOG_ERROR, "Could not open stats file %s\n", args);

Here you can use strerr(errno) for improving feedback:
            av_log(ctx, AV_LOG_ERROR, "Could not open stats file %s: %s\n", args, strerror(errno);

> +            return AVERROR(EINVAL);
> +        }
> +    }
> +
> +    return 0;
> +}
> +
> +static av_cold void uninit(AVFilterContext *ctx)
> +{
> +    PSNRContext *psnr_context = ctx->priv;
> +
> +    av_log(ctx, AV_LOG_INFO, "PSNR average:%0.2fdB min:%0.2fdB max:%0.2fdB\n",
> +                                psnr(psnr_context->mse, psnr_context->nb_frames, 255),
> +                                psnr(psnr_context->max_mse, 1, 255),
> +                                psnr(psnr_context->min_mse, 1, 255));
> +
> +    if (psnr_context->picref) {
> +        avfilter_unref_buffer(psnr_context->picref);
> +        psnr_context->picref = NULL;
> +    }

> +    if (psnr_context->line1)
> +        av_free(psnr_context->line1);
> +    if (psnr_context->line2)
> +        av_free(psnr_context->line2);

av_freep

> +    if (psnr_context->vstats_file)
> +        fclose(psnr_context->vstats_file);
> +}
> +
> +static int config_input_ref(AVFilterLink *inlink)
> +{
> +    AVFilterContext *ctx  = inlink->dst;
> +    PSNRContext *psnr_context = ctx->priv;
> +
> +    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/or heigth of input videos are different, could not calculate PSNR\n");
> +        return AVERROR(EINVAL);
> +    }
> +    if (ctx->inputs[0]->format != ctx->inputs[1]->format) {
> +        av_log(ctx, AV_LOG_ERROR, "Input filters have different pixel formats, could not calculate PSNR\n");
> +        return AVERROR(EINVAL);
> +    }
> +
> +    if (!(psnr_context->line1 = av_malloc(sizeof(*psnr_context->line1) * inlink->w)) || !(psnr_context->line2 = av_malloc(sizeof(*psnr_context->line2) * inlink->w)))
> +        return AVERROR(ENOMEM);
> +

> +    av_log(ctx, AV_LOG_INFO, "PSNR filter started correctly\n");

please remove this

> +
> +    return 0;
> +}
> +

> +static int config_output(AVFilterLink *outlink)
> +{
> +    AVFilterContext *ctx = outlink->src;
> +
> +    outlink->time_base = outlink->src->inputs[0]->time_base;
> +    outlink->w = ctx->inputs[0]->w;
> +    outlink->h = ctx->inputs[0]->h;
> +    return 0;
> +}

This should not be required IIRC.

> +
> +static int query_formats(AVFilterContext *ctx)
> +{
> +    static const enum PixelFormat pix_fmts[] = {
> +        PIX_FMT_YUV444P,  PIX_FMT_YUV422P,  PIX_FMT_YUV420P,
> +        PIX_FMT_YUV411P,  PIX_FMT_YUV410P,
> +        PIX_FMT_YUVJ444P, PIX_FMT_YUVJ422P, PIX_FMT_YUVJ420P,
> +        PIX_FMT_YUV440P,  PIX_FMT_YUVJ440P
> +    };
> +
> +    avfilter_set_common_formats(ctx, avfilter_make_format_list(pix_fmts));
> +
> +    return 0;

Would be feasible to extend the filter to all formats? compute_images_mse()
should be generic enough, and you would only need to set the max
value. Anyway this should not block the patch.

> +}
> +
> +static void start_frame(AVFilterLink *inlink, AVFilterBufferRef *inpicref)
> +{
> +    AVFilterBufferRef *outpicref = avfilter_ref_buffer(inpicref, ~0);
> +    AVFilterContext *ctx = inlink->dst;
> +    PSNRContext *psnr_context = ctx->priv;
> +
> +    inlink->dst->outputs[0]->out_buf = outpicref;
> +    outpicref->pts = av_rescale_q(inpicref->pts, ctx->inputs[0]->time_base, ctx->outputs[0]->time_base);
> +
> +    if (psnr_context->picref) {
> +        avfilter_unref_buffer(psnr_context->picref);
> +        psnr_context->picref = NULL;
> +    }
> +    avfilter_request_frame(ctx->inputs[1]);
> +
> +    avfilter_start_frame(inlink->dst->outputs[0], outpicref);
> +}
> +
> +static void start_frame_ref(AVFilterLink *inlink, AVFilterBufferRef *inpicref)
> +{
> +    AVFilterContext *ctx = inlink->dst;
> +    PSNRContext *psnr_context = ctx->priv;
> +
> +    psnr_context->picref = inpicref;
> +    psnr_context->picref->pts = av_rescale_q(inpicref->pts, ctx->inputs[1]->time_base, ctx->outputs[0]->time_base);
> +}
> +static void end_frame(AVFilterLink *inlink)
> +{
> +    AVFilterContext *ctx = inlink->dst;
> +    PSNRContext *psnr_context = ctx->priv;
> +    AVFilterLink *outlink = ctx->outputs[0];
> +    AVFilterBufferRef *outpic = outlink->out_buf;
> +    AVFilterBufferRef *ref = psnr_context->picref;
> +    double mse[4];
> +    double mse_t = 0;
> +
> +    if (psnr_context->picref) {
> +        av_compute_images_mse((const uint8_t **) outpic->data, (const uint8_t **) ref->data,
> +                              outpic->linesize, outpic->video->w, outpic->video->h, 
> +                              &av_pix_fmt_descriptors[inlink->format], mse, psnr_context->line1,
> +                              psnr_context->line2);
> +
> +        for(int j = 0; j < av_pix_fmt_descriptors[inlink->format].nb_components; j++) {
> +            mse_t += mse[j];
> +        }
> +        mse_t /= av_pix_fmt_descriptors[inlink->format].nb_components;
> +
> +        if (psnr_context->min_mse == -1) {
> +            psnr_context->min_mse = mse_t;
> +            psnr_context->max_mse = mse_t;
> +        }
> +        if (psnr_context->min_mse > mse_t)
> +            psnr_context->min_mse = mse_t;
> +        if (psnr_context->max_mse < mse_t)
> +            psnr_context->max_mse = mse_t;
> +
> +        psnr_context->mse += mse_t;
> +        psnr_context->nb_frames++;
> +
> +        if (psnr_context->vstats_file) {
> +            fprintf(psnr_context->vstats_file, "Frame:%d Y:%0.2fdB Cb:%0.2fdB Cr:%0.2fdB PSNR:%0.2fdB\n",
> +                                                psnr_context->nb_frames,
> +                                                psnr(mse[0], 1, 255),
> +                                                psnr(mse[1], 1, 255),
> +                                                psnr(mse[2], 1, 255),
> +                                                psnr(mse_t, 1, 255));
> +        }
> +    }
> +
> +    avfilter_end_frame(outlink);
> +    avfilter_unref_buffer(inlink->cur_buf);
> +}
> +
> +static void null_draw_slice(AVFilterLink *inlink, int y, int h, int slice_dir) { }
> +
> +static void null_end_frame(AVFilterLink *inlink) { }
> +
> +AVFilter avfilter_vf_psnr = {
> +    .name           = "psnr",
> +    .description    = NULL_IF_CONFIG_SMALL("Calculates the PSNR given two input files in YUV format."),
> +
> +    .init           = init,
> +    .uninit         = uninit,
> +    .query_formats  = query_formats,
> +
> +    .priv_size      = sizeof(PSNRContext),
> +
> +    .inputs         = (AVFilterPad[]) {{ .name            = "main",
> +                                         .type            = AVMEDIA_TYPE_VIDEO,
> +                                         .start_frame     = start_frame,
> +                                         .draw_slice      = null_draw_slice,
> +                                         .end_frame       = end_frame,
> +                                         .min_perms       = AV_PERM_READ, },
> +                                         { .name            = "ref",
> +                                         .type            = AVMEDIA_TYPE_VIDEO,
> +                                         .start_frame     = start_frame_ref,
> +                                         .config_props    = config_input_ref,
> +                                         .draw_slice      = null_draw_slice,
> +                                         .end_frame       = null_end_frame,
> +                                         .min_perms       = AV_PERM_READ, },
> +                                         { .name = NULL}},
> +    .outputs        = (AVFilterPad[]) {{ .name            = "default",
> +                                         .type            = AVMEDIA_TYPE_VIDEO,
> +                                         .config_props    = config_output, },
> +                                         { .name = NULL}},
> +};

Please also add a corresponding section in doc/filters.texi.
-- 
FFmpeg = F***ing and Formidable Mere Puristic Elected Gadget


More information about the ffmpeg-devel mailing list