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/opt.h"
29 #include "libavutil/pixdesc.h"
31 #include "dualinput.h"
32 #include "drawutils.h"
37 typedef struct PSNRContext {
39 FFDualInputContext dinput;
40 double mse, min_mse, max_mse;
44 int max[4], average_max;
52 void (*compute_mse)(struct PSNRContext *s,
53 const uint8_t *m[4], const int ml[4],
54 const uint8_t *r[4], const int rl[4],
55 int w, int h, double mse[4]);
58 #define OFFSET(x) offsetof(PSNRContext, x)
59 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
61 static const AVOption psnr_options[] = {
62 {"stats_file", "Set file where to store per-frame difference information", OFFSET(stats_file_str), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, FLAGS },
63 {"f", "Set file where to store per-frame difference information", OFFSET(stats_file_str), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, FLAGS },
67 AVFILTER_DEFINE_CLASS(psnr);
69 static inline unsigned pow2(unsigned base)
74 static inline double get_psnr(double mse, uint64_t nb_frames, int max)
76 return 10.0 * log(pow2(max) / (mse / nb_frames)) / log(10.0);
80 void compute_images_mse(PSNRContext *s,
81 const uint8_t *main_data[4], const int main_linesizes[4],
82 const uint8_t *ref_data[4], const int ref_linesizes[4],
83 int w, int h, double mse[4])
87 for (c = 0; c < s->nb_components; c++) {
88 const int outw = s->planewidth[c];
89 const int outh = s->planeheight[c];
90 const uint8_t *main_line = main_data[c];
91 const uint8_t *ref_line = ref_data[c];
92 const int ref_linesize = ref_linesizes[c];
93 const int main_linesize = main_linesizes[c];
96 for (i = 0; i < outh; i++) {
98 for (j = 0; j < outw; j++)
99 m2 += pow2(main_line[j] - ref_line[j]);
101 ref_line += ref_linesize;
102 main_line += main_linesize;
104 mse[c] = m / (double)(outw * outh);
109 void compute_images_mse_16bit(PSNRContext *s,
110 const uint8_t *main_data[4], const int main_linesizes[4],
111 const uint8_t *ref_data[4], const int ref_linesizes[4],
112 int w, int h, double mse[4])
116 for (c = 0; c < s->nb_components; c++) {
117 const int outw = s->planewidth[c];
118 const int outh = s->planeheight[c];
119 const uint16_t *main_line = (uint16_t *)main_data[c];
120 const uint16_t *ref_line = (uint16_t *)ref_data[c];
121 const int ref_linesize = ref_linesizes[c] / 2;
122 const int main_linesize = main_linesizes[c] / 2;
125 for (i = 0; i < outh; i++) {
126 for (j = 0; j < outw; j++)
127 m += pow2(main_line[j] - ref_line[j]);
128 ref_line += ref_linesize;
129 main_line += main_linesize;
131 mse[c] = m / (double)(outw * outh);
135 static void set_meta(AVDictionary **metadata, const char *key, char comp, float d)
138 snprintf(value, sizeof(value), "%0.2f", d);
141 snprintf(key2, sizeof(key2), "%s%c", key, comp);
142 av_dict_set(metadata, key2, value, 0);
144 av_dict_set(metadata, key, value, 0);
148 static AVFrame *do_psnr(AVFilterContext *ctx, AVFrame *main,
151 PSNRContext *s = ctx->priv;
152 double comp_mse[4], mse = 0;
154 AVDictionary **metadata = avpriv_frame_get_metadatap(main);
156 s->compute_mse(s, (const uint8_t **)main->data, main->linesize,
157 (const uint8_t **)ref->data, ref->linesize,
158 main->width, main->height, comp_mse);
160 for (j = 0; j < s->nb_components; j++)
162 mse /= s->nb_components;
164 s->min_mse = FFMIN(s->min_mse, mse);
165 s->max_mse = FFMAX(s->max_mse, mse);
170 for (j = 0; j < s->nb_components; j++) {
171 c = s->is_rgb ? s->rgba_map[j] : j;
172 set_meta(metadata, "lavfi.psnr.mse.", s->comps[j], comp_mse[c]);
173 set_meta(metadata, "lavfi.psnr.mse_avg", 0, mse);
174 set_meta(metadata, "lavfi.psnr.psnr.", s->comps[j], get_psnr(comp_mse[c], 1, s->max[c]));
175 set_meta(metadata, "lavfi.psnr.psnr_avg", 0, get_psnr(mse, 1, s->average_max));
179 fprintf(s->stats_file, "n:%"PRId64" mse_avg:%0.2f ", s->nb_frames, mse);
180 for (j = 0; j < s->nb_components; j++) {
181 c = s->is_rgb ? s->rgba_map[j] : j;
182 fprintf(s->stats_file, "mse_%c:%0.2f ", s->comps[j], comp_mse[c]);
184 for (j = 0; j < s->nb_components; j++) {
185 c = s->is_rgb ? s->rgba_map[j] : j;
186 fprintf(s->stats_file, "psnr_%c:%0.2f ", s->comps[j],
187 get_psnr(comp_mse[c], 1, s->max[c]));
189 fprintf(s->stats_file, "\n");
195 static av_cold int init(AVFilterContext *ctx)
197 PSNRContext *s = ctx->priv;
199 s->min_mse = +INFINITY;
200 s->max_mse = -INFINITY;
202 if (s->stats_file_str) {
203 s->stats_file = fopen(s->stats_file_str, "w");
204 if (!s->stats_file) {
205 int err = AVERROR(errno);
207 av_strerror(err, buf, sizeof(buf));
208 av_log(ctx, AV_LOG_ERROR, "Could not open stats file %s: %s\n",
209 s->stats_file_str, buf);
214 s->dinput.process = do_psnr;
218 static int query_formats(AVFilterContext *ctx)
220 static const enum AVPixelFormat pix_fmts[] = {
221 AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY16,
222 #define PF_NOALPHA(suf) AV_PIX_FMT_YUV420##suf, AV_PIX_FMT_YUV422##suf, AV_PIX_FMT_YUV444##suf
223 #define PF_ALPHA(suf) AV_PIX_FMT_YUVA420##suf, AV_PIX_FMT_YUVA422##suf, AV_PIX_FMT_YUVA444##suf
224 #define PF(suf) PF_NOALPHA(suf), PF_ALPHA(suf)
225 PF(P), PF(P9), PF(P10), PF_NOALPHA(P12), PF_NOALPHA(P14), PF(P16),
226 AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
227 AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P,
228 AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_YUVJ444P,
229 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
230 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
231 AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP16,
235 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
237 return AVERROR(ENOMEM);
238 return ff_set_common_formats(ctx, fmts_list);
241 static int config_input_ref(AVFilterLink *inlink)
243 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
244 AVFilterContext *ctx = inlink->dst;
245 PSNRContext *s = ctx->priv;
248 s->nb_components = desc->nb_components;
249 if (ctx->inputs[0]->w != ctx->inputs[1]->w ||
250 ctx->inputs[0]->h != ctx->inputs[1]->h) {
251 av_log(ctx, AV_LOG_ERROR, "Width and height of input videos must be same.\n");
252 return AVERROR(EINVAL);
254 if (ctx->inputs[0]->format != ctx->inputs[1]->format) {
255 av_log(ctx, AV_LOG_ERROR, "Inputs must be of same pixel format.\n");
256 return AVERROR(EINVAL);
259 switch (inlink->format) {
260 case AV_PIX_FMT_GRAY8:
261 case AV_PIX_FMT_GRAY16:
262 case AV_PIX_FMT_GBRP:
263 case AV_PIX_FMT_GBRP9:
264 case AV_PIX_FMT_GBRP10:
265 case AV_PIX_FMT_GBRP12:
266 case AV_PIX_FMT_GBRP14:
267 case AV_PIX_FMT_GBRP16:
268 case AV_PIX_FMT_GBRAP:
269 case AV_PIX_FMT_GBRAP16:
270 case AV_PIX_FMT_YUVJ411P:
271 case AV_PIX_FMT_YUVJ420P:
272 case AV_PIX_FMT_YUVJ422P:
273 case AV_PIX_FMT_YUVJ440P:
274 case AV_PIX_FMT_YUVJ444P:
275 s->max[0] = (1 << (desc->comp[0].depth_minus1 + 1)) - 1;
276 s->max[1] = (1 << (desc->comp[1].depth_minus1 + 1)) - 1;
277 s->max[2] = (1 << (desc->comp[2].depth_minus1 + 1)) - 1;
278 s->max[3] = (1 << (desc->comp[3].depth_minus1 + 1)) - 1;
281 s->max[0] = 235 * (1 << (desc->comp[0].depth_minus1 - 7));
282 s->max[1] = 240 * (1 << (desc->comp[1].depth_minus1 - 7));
283 s->max[2] = 240 * (1 << (desc->comp[2].depth_minus1 - 7));
284 s->max[3] = (1 << (desc->comp[3].depth_minus1 + 1)) - 1;
287 s->is_rgb = ff_fill_rgba_map(s->rgba_map, inlink->format) >= 0;
288 s->comps[0] = s->is_rgb ? 'r' : 'y' ;
289 s->comps[1] = s->is_rgb ? 'g' : 'u' ;
290 s->comps[2] = s->is_rgb ? 'b' : 'v' ;
293 for (j = 0; j < s->nb_components; j++)
294 s->average_max += s->max[j];
295 s->average_max /= s->nb_components;
297 s->planeheight[1] = s->planeheight[2] = FF_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
298 s->planeheight[0] = s->planeheight[3] = inlink->h;
299 s->planewidth[1] = s->planewidth[2] = FF_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
300 s->planewidth[0] = s->planewidth[3] = inlink->w;
302 s->compute_mse = desc->comp[0].depth_minus1 > 7 ? compute_images_mse_16bit : compute_images_mse;
307 static int config_output(AVFilterLink *outlink)
309 AVFilterContext *ctx = outlink->src;
310 PSNRContext *s = ctx->priv;
311 AVFilterLink *mainlink = ctx->inputs[0];
314 outlink->w = mainlink->w;
315 outlink->h = mainlink->h;
316 outlink->time_base = mainlink->time_base;
317 outlink->sample_aspect_ratio = mainlink->sample_aspect_ratio;
318 outlink->frame_rate = mainlink->frame_rate;
319 if ((ret = ff_dualinput_init(ctx, &s->dinput)) < 0)
325 static int filter_frame(AVFilterLink *inlink, AVFrame *inpicref)
327 PSNRContext *s = inlink->dst->priv;
328 return ff_dualinput_filter_frame(&s->dinput, inlink, inpicref);
331 static int request_frame(AVFilterLink *outlink)
333 PSNRContext *s = outlink->src->priv;
334 return ff_dualinput_request_frame(&s->dinput, outlink);
337 static av_cold void uninit(AVFilterContext *ctx)
339 PSNRContext *s = ctx->priv;
341 if (s->nb_frames > 0) {
342 av_log(ctx, AV_LOG_INFO, "PSNR average:%0.2f min:%0.2f max:%0.2f\n",
343 get_psnr(s->mse, s->nb_frames, s->average_max),
344 get_psnr(s->max_mse, 1, s->average_max),
345 get_psnr(s->min_mse, 1, s->average_max));
348 ff_dualinput_uninit(&s->dinput);
351 fclose(s->stats_file);
354 static const AVFilterPad psnr_inputs[] = {
357 .type = AVMEDIA_TYPE_VIDEO,
358 .filter_frame = filter_frame,
361 .type = AVMEDIA_TYPE_VIDEO,
362 .filter_frame = filter_frame,
363 .config_props = config_input_ref,
368 static const AVFilterPad psnr_outputs[] = {
371 .type = AVMEDIA_TYPE_VIDEO,
372 .config_props = config_output,
373 .request_frame = request_frame,
378 AVFilter ff_vf_psnr = {
380 .description = NULL_IF_CONFIG_SMALL("Calculate the PSNR between two video streams."),
383 .query_formats = query_formats,
384 .priv_size = sizeof(PSNRContext),
385 .priv_class = &psnr_class,
386 .inputs = psnr_inputs,
387 .outputs = psnr_outputs,