2 * Copyright (c) 2011 Roger Pau Monné <roger.pau@entel.upc.edu>
3 * Copyright (c) 2011 Stefano Sabatini
4 * Copyright (c) 2013 Paul B Mahol
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * Caculate the PSNR between two input videos.
28 #include "libavutil/avstring.h"
29 #include "libavutil/opt.h"
30 #include "libavutil/pixdesc.h"
32 #include "dualinput.h"
33 #include "drawutils.h"
39 typedef struct PSNRContext {
41 FFDualInputContext dinput;
42 double mse, min_mse, max_mse, mse_comp[4];
46 int max[4], average_max;
53 double planeweight[4];
57 #define OFFSET(x) offsetof(PSNRContext, x)
58 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
60 static const AVOption psnr_options[] = {
61 {"stats_file", "Set file where to store per-frame difference information", OFFSET(stats_file_str), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, FLAGS },
62 {"f", "Set file where to store per-frame difference information", OFFSET(stats_file_str), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, FLAGS },
66 AVFILTER_DEFINE_CLASS(psnr);
68 static inline unsigned pow2(unsigned base)
73 static inline double get_psnr(double mse, uint64_t nb_frames, int max)
75 return 10.0 * log(pow2(max) / (mse / nb_frames)) / log(10.0);
78 static uint64_t sse_line_8bit(const uint8_t *main_line, const uint8_t *ref_line, int outw)
83 for (j = 0; j < outw; j++)
84 m2 += pow2(main_line[j] - ref_line[j]);
89 static uint64_t sse_line_16bit(const uint8_t *_main_line, const uint8_t *_ref_line, int outw)
93 const uint16_t *main_line = (const uint16_t *) _main_line;
94 const uint16_t *ref_line = (const uint16_t *) _ref_line;
96 for (j = 0; j < outw; j++)
97 m2 += pow2(main_line[j] - ref_line[j]);
103 void compute_images_mse(PSNRContext *s,
104 const uint8_t *main_data[4], const int main_linesizes[4],
105 const uint8_t *ref_data[4], const int ref_linesizes[4],
106 int w, int h, double mse[4])
110 for (c = 0; c < s->nb_components; c++) {
111 const int outw = s->planewidth[c];
112 const int outh = s->planeheight[c];
113 const uint8_t *main_line = main_data[c];
114 const uint8_t *ref_line = ref_data[c];
115 const int ref_linesize = ref_linesizes[c];
116 const int main_linesize = main_linesizes[c];
118 for (i = 0; i < outh; i++) {
119 m += s->dsp.sse_line(main_line, ref_line, outw);
120 ref_line += ref_linesize;
121 main_line += main_linesize;
123 mse[c] = m / (double)(outw * outh);
127 static void set_meta(AVDictionary **metadata, const char *key, char comp, float d)
130 snprintf(value, sizeof(value), "%0.2f", d);
133 snprintf(key2, sizeof(key2), "%s%c", key, comp);
134 av_dict_set(metadata, key2, value, 0);
136 av_dict_set(metadata, key, value, 0);
140 static AVFrame *do_psnr(AVFilterContext *ctx, AVFrame *main,
143 PSNRContext *s = ctx->priv;
144 double comp_mse[4], mse = 0;
146 AVDictionary **metadata = avpriv_frame_get_metadatap(main);
148 compute_images_mse(s, (const uint8_t **)main->data, main->linesize,
149 (const uint8_t **)ref->data, ref->linesize,
150 main->width, main->height, comp_mse);
152 for (j = 0; j < s->nb_components; j++)
153 mse += comp_mse[j] * s->planeweight[j];
155 s->min_mse = FFMIN(s->min_mse, mse);
156 s->max_mse = FFMAX(s->max_mse, mse);
159 for (j = 0; j < s->nb_components; j++)
160 s->mse_comp[j] += comp_mse[j];
163 for (j = 0; j < s->nb_components; j++) {
164 c = s->is_rgb ? s->rgba_map[j] : j;
165 set_meta(metadata, "lavfi.psnr.mse.", s->comps[j], comp_mse[c]);
166 set_meta(metadata, "lavfi.psnr.psnr.", s->comps[j], get_psnr(comp_mse[c], 1, s->max[c]));
168 set_meta(metadata, "lavfi.psnr.mse_avg", 0, mse);
169 set_meta(metadata, "lavfi.psnr.psnr_avg", 0, get_psnr(mse, 1, s->average_max));
172 fprintf(s->stats_file, "n:%"PRId64" mse_avg:%0.2f ", s->nb_frames, mse);
173 for (j = 0; j < s->nb_components; j++) {
174 c = s->is_rgb ? s->rgba_map[j] : j;
175 fprintf(s->stats_file, "mse_%c:%0.2f ", s->comps[j], comp_mse[c]);
177 for (j = 0; j < s->nb_components; j++) {
178 c = s->is_rgb ? s->rgba_map[j] : j;
179 fprintf(s->stats_file, "psnr_%c:%0.2f ", s->comps[j],
180 get_psnr(comp_mse[c], 1, s->max[c]));
182 fprintf(s->stats_file, "\n");
188 static av_cold int init(AVFilterContext *ctx)
190 PSNRContext *s = ctx->priv;
192 s->min_mse = +INFINITY;
193 s->max_mse = -INFINITY;
195 if (s->stats_file_str) {
196 s->stats_file = fopen(s->stats_file_str, "w");
197 if (!s->stats_file) {
198 int err = AVERROR(errno);
200 av_strerror(err, buf, sizeof(buf));
201 av_log(ctx, AV_LOG_ERROR, "Could not open stats file %s: %s\n",
202 s->stats_file_str, buf);
207 s->dinput.process = do_psnr;
211 static int query_formats(AVFilterContext *ctx)
213 static const enum AVPixelFormat pix_fmts[] = {
214 AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY16,
215 #define PF_NOALPHA(suf) AV_PIX_FMT_YUV420##suf, AV_PIX_FMT_YUV422##suf, AV_PIX_FMT_YUV444##suf
216 #define PF_ALPHA(suf) AV_PIX_FMT_YUVA420##suf, AV_PIX_FMT_YUVA422##suf, AV_PIX_FMT_YUVA444##suf
217 #define PF(suf) PF_NOALPHA(suf), PF_ALPHA(suf)
218 PF(P), PF(P9), PF(P10), PF_NOALPHA(P12), PF_NOALPHA(P14), PF(P16),
219 AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
220 AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P,
221 AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_YUVJ444P,
222 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
223 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
224 AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP16,
228 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
230 return AVERROR(ENOMEM);
231 return ff_set_common_formats(ctx, fmts_list);
234 static int config_input_ref(AVFilterLink *inlink)
236 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
237 AVFilterContext *ctx = inlink->dst;
238 PSNRContext *s = ctx->priv;
242 s->nb_components = desc->nb_components;
243 if (ctx->inputs[0]->w != ctx->inputs[1]->w ||
244 ctx->inputs[0]->h != ctx->inputs[1]->h) {
245 av_log(ctx, AV_LOG_ERROR, "Width and height of input videos must be same.\n");
246 return AVERROR(EINVAL);
248 if (ctx->inputs[0]->format != ctx->inputs[1]->format) {
249 av_log(ctx, AV_LOG_ERROR, "Inputs must be of same pixel format.\n");
250 return AVERROR(EINVAL);
253 s->max[0] = (1 << (desc->comp[0].depth_minus1 + 1)) - 1;
254 s->max[1] = (1 << (desc->comp[1].depth_minus1 + 1)) - 1;
255 s->max[2] = (1 << (desc->comp[2].depth_minus1 + 1)) - 1;
256 s->max[3] = (1 << (desc->comp[3].depth_minus1 + 1)) - 1;
258 s->is_rgb = ff_fill_rgba_map(s->rgba_map, inlink->format) >= 0;
259 s->comps[0] = s->is_rgb ? 'r' : 'y' ;
260 s->comps[1] = s->is_rgb ? 'g' : 'u' ;
261 s->comps[2] = s->is_rgb ? 'b' : 'v' ;
264 s->planeheight[1] = s->planeheight[2] = FF_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
265 s->planeheight[0] = s->planeheight[3] = inlink->h;
266 s->planewidth[1] = s->planewidth[2] = FF_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
267 s->planewidth[0] = s->planewidth[3] = inlink->w;
269 for (j = 0; j < s->nb_components; j++)
270 sum += s->planeheight[j] * s->planewidth[j];
271 for (j = 0; j < s->nb_components; j++) {
272 s->planeweight[j] = (double) s->planeheight[j] * s->planewidth[j] / sum;
273 s->average_max += s->max[j] * s->planeweight[j];
276 s->dsp.sse_line = desc->comp[0].depth_minus1 > 7 ? sse_line_16bit : sse_line_8bit;
278 ff_psnr_init_x86(&s->dsp, desc->comp[0].depth_minus1 + 1);
283 static int config_output(AVFilterLink *outlink)
285 AVFilterContext *ctx = outlink->src;
286 PSNRContext *s = ctx->priv;
287 AVFilterLink *mainlink = ctx->inputs[0];
290 outlink->w = mainlink->w;
291 outlink->h = mainlink->h;
292 outlink->time_base = mainlink->time_base;
293 outlink->sample_aspect_ratio = mainlink->sample_aspect_ratio;
294 outlink->frame_rate = mainlink->frame_rate;
295 if ((ret = ff_dualinput_init(ctx, &s->dinput)) < 0)
301 static int filter_frame(AVFilterLink *inlink, AVFrame *inpicref)
303 PSNRContext *s = inlink->dst->priv;
304 return ff_dualinput_filter_frame(&s->dinput, inlink, inpicref);
307 static int request_frame(AVFilterLink *outlink)
309 PSNRContext *s = outlink->src->priv;
310 return ff_dualinput_request_frame(&s->dinput, outlink);
313 static av_cold void uninit(AVFilterContext *ctx)
315 PSNRContext *s = ctx->priv;
317 if (s->nb_frames > 0) {
322 for (j = 0; j < s->nb_components; j++) {
323 int c = s->is_rgb ? s->rgba_map[j] : j;
324 av_strlcatf(buf, sizeof(buf), " %c:%0.2f", s->comps[j],
325 get_psnr(s->mse_comp[c], s->nb_frames, s->max[c]));
327 av_log(ctx, AV_LOG_INFO, "PSNR%s average:%0.2f min:%0.2f max:%0.2f\n",
329 get_psnr(s->mse, s->nb_frames, s->average_max),
330 get_psnr(s->max_mse, 1, s->average_max),
331 get_psnr(s->min_mse, 1, s->average_max));
334 ff_dualinput_uninit(&s->dinput);
337 fclose(s->stats_file);
340 static const AVFilterPad psnr_inputs[] = {
343 .type = AVMEDIA_TYPE_VIDEO,
344 .filter_frame = filter_frame,
347 .type = AVMEDIA_TYPE_VIDEO,
348 .filter_frame = filter_frame,
349 .config_props = config_input_ref,
354 static const AVFilterPad psnr_outputs[] = {
357 .type = AVMEDIA_TYPE_VIDEO,
358 .config_props = config_output,
359 .request_frame = request_frame,
364 AVFilter ff_vf_psnr = {
366 .description = NULL_IF_CONFIG_SMALL("Calculate the PSNR between two video streams."),
369 .query_formats = query_formats,
370 .priv_size = sizeof(PSNRContext),
371 .priv_class = &psnr_class,
372 .inputs = psnr_inputs,
373 .outputs = psnr_outputs,