FFmpeg coverage


Directory: ../../../ffmpeg/
File: src/libavfilter/vf_xpsnr.c
Date: 2025-01-20 09:27:23
Exec Total Coverage
Lines: 265 357 74.2%
Functions: 12 16 75.0%
Branches: 156 250 62.4%

Line Branch Exec Source
1 /*
2 * Copyright (c) 2024 Christian R. Helmrich
3 * Copyright (c) 2024 Christian Lehmann
4 * Copyright (c) 2024 Christian Stoffers
5 *
6 * This file is part of FFmpeg.
7 *
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.
12 *
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.
17 *
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
21 */
22
23 /**
24 * @file
25 * Calculate the extended perceptually weighted PSNR (XPSNR) between two input videos.
26 *
27 * Authors: Christian Helmrich, Lehmann, and Stoffers, Fraunhofer HHI, Berlin, Germany
28 */
29
30 #include "libavutil/avstring.h"
31 #include "libavutil/file_open.h"
32 #include "libavutil/mem.h"
33 #include "libavutil/opt.h"
34 #include "libavutil/pixdesc.h"
35 #include "avfilter.h"
36 #include "drawutils.h"
37 #include "filters.h"
38 #include "framesync.h"
39 #include "psnr.h"
40 #include "xpsnr.h"
41
42 /* XPSNR structure definition */
43
44 typedef struct XPSNRContext {
45 /* required basic variables */
46 const AVClass *class;
47 int bpp; /* unpacked */
48 int depth; /* packed */
49 char comps[4];
50 int num_comps;
51 uint64_t num_frames_64;
52 unsigned frame_rate;
53 FFFrameSync fs;
54 int line_sizes[4];
55 int plane_height[4];
56 int plane_width[4];
57 uint8_t rgba_map[4];
58 FILE *stats_file;
59 char *stats_file_str;
60 /* XPSNR specific variables */
61 double *sse_luma;
62 double *weights;
63 int16_t *buf_org_m1;
64 int16_t *buf_org_m2;
65 int16_t *buf_org [3];
66 int16_t *buf_rec [3];
67 uint64_t max_error_64;
68 double sum_wdist [3];
69 double sum_xpsnr [3];
70 int and_is_inf[3];
71 int is_rgb;
72 XPSNRDSPContext dsp;
73 PSNRDSPContext pdsp;
74 } XPSNRContext;
75
76 /* required macro definitions */
77
78 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM
79 #define OFFSET(x) offsetof(XPSNRContext, x)
80 #define XPSNR_GAMMA 2
81
82 static const AVOption xpsnr_options[] = {
83 {"stats_file", "Set file where to store per-frame XPSNR information", OFFSET(stats_file_str), AV_OPT_TYPE_STRING, {.str = NULL}, 0, 0, FLAGS},
84 {"f", "Set file where to store per-frame XPSNR information", OFFSET(stats_file_str), AV_OPT_TYPE_STRING, {.str = NULL}, 0, 0, FLAGS},
85 { NULL }
86 };
87
88
0/2
✗ Branch 0 not taken.
✗ Branch 1 not taken.
8 FRAMESYNC_DEFINE_CLASS(xpsnr, XPSNRContext, fs);
89
90 /* XPSNR function definitions */
91
92 static uint64_t highds(const int x_act, const int y_act, const int w_act, const int h_act, const int16_t *o_m0, const int o)
93 {
94 uint64_t sa_act = 0;
95
96 for (int y = y_act; y < h_act; y += 2) {
97 for (int x = x_act; x < w_act; x += 2) {
98 const int f = 12 * ((int)o_m0[ y *o + x ] + (int)o_m0[ y *o + x+1] + (int)o_m0[(y+1)*o + x ] + (int)o_m0[(y+1)*o + x+1])
99 - 3 * ((int)o_m0[(y-1)*o + x ] + (int)o_m0[(y-1)*o + x+1] + (int)o_m0[(y+2)*o + x ] + (int)o_m0[(y+2)*o + x+1])
100 - 3 * ((int)o_m0[ y *o + x-1] + (int)o_m0[ y *o + x+2] + (int)o_m0[(y+1)*o + x-1] + (int)o_m0[(y+1)*o + x+2])
101 - 2 * ((int)o_m0[(y-1)*o + x-1] + (int)o_m0[(y-1)*o + x+2] + (int)o_m0[(y+2)*o + x-1] + (int)o_m0[(y+2)*o + x+2])
102 - ((int)o_m0[(y-2)*o + x-1] + (int)o_m0[(y-2)*o + x ] + (int)o_m0[(y-2)*o + x+1] + (int)o_m0[(y-2)*o + x+2]
103 + (int)o_m0[(y+3)*o + x-1] + (int)o_m0[(y+3)*o + x ] + (int)o_m0[(y+3)*o + x+1] + (int)o_m0[(y+3)*o + x+2]
104 + (int)o_m0[(y-1)*o + x-2] + (int)o_m0[ y *o + x-2] + (int)o_m0[(y+1)*o + x-2] + (int)o_m0[(y+2)*o + x-2]
105 + (int)o_m0[(y-1)*o + x+3] + (int)o_m0[ y *o + x+3] + (int)o_m0[(y+1)*o + x+3] + (int)o_m0[(y+2)*o + x+3]);
106 sa_act += (uint64_t) abs(f);
107 }
108 }
109 return sa_act;
110 }
111
112 static uint64_t diff1st(const uint32_t w_act, const uint32_t h_act, const int16_t *o_m0, int16_t *o_m1, const int o)
113 {
114 uint64_t ta_act = 0;
115
116 for (uint32_t y = 0; y < h_act; y += 2) {
117 for (uint32_t x = 0; x < w_act; x += 2) {
118 const int t = (int)o_m0[y*o + x] + (int)o_m0[y*o + x+1] + (int)o_m0[(y+1)*o + x] + (int)o_m0[(y+1)*o + x+1]
119 - ((int)o_m1[y*o + x] + (int)o_m1[y*o + x+1] + (int)o_m1[(y+1)*o + x] + (int)o_m1[(y+1)*o + x+1]);
120 ta_act += (uint64_t) abs(t);
121 o_m1[y*o + x ] = o_m0[y*o + x ]; o_m1[(y+1)*o + x ] = o_m0[(y+1)*o + x ];
122 o_m1[y*o + x+1] = o_m0[y*o + x+1]; o_m1[(y+1)*o + x+1] = o_m0[(y+1)*o + x+1];
123 }
124 }
125 return (ta_act * XPSNR_GAMMA);
126 }
127
128 static uint64_t diff2nd(const uint32_t w_act, const uint32_t h_act, const int16_t *o_m0, int16_t *o_m1, int16_t *o_m2, const int o)
129 {
130 uint64_t ta_act = 0;
131
132 for (uint32_t y = 0; y < h_act; y += 2) {
133 for (uint32_t x = 0; x < w_act; x += 2) {
134 const int t = (int)o_m0[y*o + x] + (int)o_m0[y*o + x+1] + (int)o_m0[(y+1)*o + x] + (int)o_m0[(y+1)*o + x+1]
135 - 2 * ((int)o_m1[y*o + x] + (int)o_m1[y*o + x+1] + (int)o_m1[(y+1)*o + x] + (int)o_m1[(y+1)*o + x+1])
136 + (int)o_m2[y*o + x] + (int)o_m2[y*o + x+1] + (int)o_m2[(y+1)*o + x] + (int)o_m2[(y+1)*o + x+1];
137 ta_act += (uint64_t) abs(t);
138 o_m2[y*o + x ] = o_m1[y*o + x ]; o_m2[(y+1)*o + x ] = o_m1[(y+1)*o + x ];
139 o_m2[y*o + x+1] = o_m1[y*o + x+1]; o_m2[(y+1)*o + x+1] = o_m1[(y+1)*o + x+1];
140 o_m1[y*o + x ] = o_m0[y*o + x ]; o_m1[(y+1)*o + x ] = o_m0[(y+1)*o + x ];
141 o_m1[y*o + x+1] = o_m0[y*o + x+1]; o_m1[(y+1)*o + x+1] = o_m0[(y+1)*o + x+1];
142 }
143 }
144 return (ta_act * XPSNR_GAMMA);
145 }
146
147 12750 static inline uint64_t calc_squared_error(XPSNRContext const *s,
148 const int16_t *blk_org, const uint32_t stride_org,
149 const int16_t *blk_rec, const uint32_t stride_rec,
150 const uint32_t block_width, const uint32_t block_height)
151 {
152 12750 uint64_t sse = 0; /* sum of squared errors */
153
154
2/2
✓ Branch 0 taken 150000 times.
✓ Branch 1 taken 12750 times.
162750 for (uint32_t y = 0; y < block_height; y++) {
155 150000 sse += s->pdsp.sse_line((const uint8_t *) blk_org, (const uint8_t *) blk_rec, (int) block_width);
156 150000 blk_org += stride_org;
157 150000 blk_rec += stride_rec;
158 }
159
160 /* return nonweighted sum of squared errors */
161 12750 return sse;
162 }
163
164 4250 static inline double calc_squared_error_and_weight (XPSNRContext const *s,
165 const int16_t *pic_org, const uint32_t stride_org,
166 int16_t *pic_org_m1, int16_t *pic_org_m2,
167 const int16_t *pic_rec, const uint32_t stride_rec,
168 const uint32_t offset_x, const uint32_t offset_y,
169 const uint32_t block_width, const uint32_t block_height,
170 const uint32_t bit_depth, const uint32_t int_frame_rate, double *ms_act)
171 {
172 4250 const int o = (int) stride_org;
173 4250 const int r = (int) stride_rec;
174 4250 const int16_t *o_m0 = pic_org + offset_y * o + offset_x;
175 4250 int16_t *o_m1 = pic_org_m1 + offset_y * o + offset_x;
176 4250 int16_t *o_m2 = pic_org_m2 + offset_y * o + offset_x;
177 4250 const int16_t *r_m0 = pic_rec + offset_y * r + offset_x;
178
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4250 times.
4250 const int b_val = (s->plane_width[0] * s->plane_height[0] > 2048 * 1152 ? 2 : 1); /* threshold is a bit more than HD resolution */
179
2/2
✓ Branch 0 taken 170 times.
✓ Branch 1 taken 4080 times.
4250 const int x_act = (offset_x > 0 ? 0 : b_val);
180
2/2
✓ Branch 0 taken 250 times.
✓ Branch 1 taken 4000 times.
4250 const int y_act = (offset_y > 0 ? 0 : b_val);
181
2/2
✓ Branch 0 taken 170 times.
✓ Branch 1 taken 4080 times.
4250 const int w_act = (offset_x + block_width < (uint32_t) s->plane_width [0] ? (int) block_width : (int) block_width - b_val);
182
2/2
✓ Branch 0 taken 250 times.
✓ Branch 1 taken 4000 times.
4250 const int h_act = (offset_y + block_height < (uint32_t) s->plane_height[0] ? (int) block_height : (int) block_height - b_val);
183
184 4250 const double sse = (double) calc_squared_error (s, o_m0, stride_org,
185 r_m0, stride_rec,
186 block_width, block_height);
187 4250 uint64_t sa_act = 0; /* spatial abs. activity */
188 4250 uint64_t ta_act = 0; /* temporal abs. activity */
189
190
2/4
✓ Branch 0 taken 4250 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 4250 times.
4250 if (w_act <= x_act || h_act <= y_act) /* small */
191 return sse;
192
193
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4250 times.
4250 if (b_val > 1) { /* highpass with downsampling */
194 if (w_act > 12)
195 sa_act = s->dsp.highds_func(x_act, y_act, w_act, h_act, o_m0, o);
196 else
197 highds(x_act, y_act, w_act, h_act, o_m0, o);
198 } else { /* <=HD highpass without downsampling */
199
2/2
✓ Branch 0 taken 49500 times.
✓ Branch 1 taken 4250 times.
53750 for (int y = y_act; y < h_act; y++) {
200
2/2
✓ Branch 0 taken 590040 times.
✓ Branch 1 taken 49500 times.
639540 for (int x = x_act; x < w_act; x++) {
201 590040 const int f = 12 * (int)o_m0[y*o + x] - 2 * ((int)o_m0[y*o + x-1] + (int)o_m0[y*o + x+1] + (int)o_m0[(y-1)*o + x] + (int)o_m0[(y+1)*o + x])
202 590040 - ((int)o_m0[(y-1)*o + x-1] + (int)o_m0[(y-1)*o + x+1] + (int)o_m0[(y+1)*o + x-1] + (int)o_m0[(y+1)*o + x+1]);
203 590040 sa_act += (uint64_t) abs(f);
204 }
205 }
206 }
207
208 /* calculate weight (average squared activity) */
209 4250 *ms_act = (double) sa_act / ((double) (w_act - x_act) * (double) (h_act - y_act));
210
211
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4250 times.
4250 if (b_val > 1) { /* highpass with downsampling */
212 if (int_frame_rate < 32) /* 1st-order diff */
213 ta_act = s->dsp.diff1st_func(block_width, block_height, o_m0, o_m1, o);
214 else /* 2nd-order diff (diff of two diffs) */
215 ta_act = s->dsp.diff2nd_func(block_width, block_height, o_m0, o_m1, o_m2, o);
216 } else { /* <=HD highpass without downsampling */
217
1/2
✓ Branch 0 taken 4250 times.
✗ Branch 1 not taken.
4250 if (int_frame_rate < 32) { /* 1st-order diff */
218
2/2
✓ Branch 0 taken 50000 times.
✓ Branch 1 taken 4250 times.
54250 for (uint32_t y = 0; y < block_height; y++) {
219
2/2
✓ Branch 0 taken 600000 times.
✓ Branch 1 taken 50000 times.
650000 for (uint32_t x = 0; x < block_width; x++) {
220 600000 const int t = (int)o_m0[y * o + x] - (int)o_m1[y * o + x];
221
222 600000 ta_act += XPSNR_GAMMA * (uint64_t) abs(t);
223 600000 o_m1[y * o + x] = o_m0[y * o + x];
224 }
225 }
226 } else { /* 2nd-order diff (diff of 2 diffs) */
227 for (uint32_t y = 0; y < block_height; y++) {
228 for (uint32_t x = 0; x < block_width; x++) {
229 const int t = (int)o_m0[y * o + x] - 2 * (int)o_m1[y * o + x] + (int)o_m2[y * o + x];
230
231 ta_act += XPSNR_GAMMA * (uint64_t) abs(t);
232 o_m2[y * o + x] = o_m1[y * o + x];
233 o_m1[y * o + x] = o_m0[y * o + x];
234 }
235 }
236 }
237 }
238
239 /* weight += mean squared temporal activity */
240 4250 *ms_act += (double) ta_act / ((double) block_width * (double) block_height);
241
242 /* lower limit, accounts for high-pass gain */
243
2/2
✓ Branch 0 taken 1813 times.
✓ Branch 1 taken 2437 times.
4250 if (*ms_act < (double) (1 << (bit_depth - 6)))
244 1813 *ms_act = (double) (1 << (bit_depth - 6));
245
246 4250 *ms_act *= *ms_act; /* since SSE is squared */
247
248 /* return nonweighted sum of squared errors */
249 4250 return sse;
250 }
251
252 36 static inline double get_avg_xpsnr (const double sqrt_wsse_val, const double sum_xpsnr_val,
253 const uint32_t image_width, const uint32_t image_height,
254 const uint64_t max_error_64, const uint64_t num_frames_64)
255 {
256
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 36 times.
36 if (num_frames_64 == 0)
257 return INFINITY;
258
259
1/2
✓ Branch 0 taken 36 times.
✗ Branch 1 not taken.
36 if (sqrt_wsse_val >= (double) num_frames_64) { /* square-mean-root average */
260 36 const double avg_dist = sqrt_wsse_val / (double) num_frames_64;
261 36 const uint64_t num64 = (uint64_t) image_width * (uint64_t) image_height * max_error_64;
262
263 36 return 10.0 * log10((double) num64 / ((double) avg_dist * (double) avg_dist));
264 }
265
266 return sum_xpsnr_val / (double) num_frames_64; /* older log-domain average */
267 }
268
269 10 static int get_wsse(AVFilterContext *ctx, int16_t **org, int16_t *org_m1,
270 int16_t *org_m2, int16_t **rec, uint64_t *const wsse64)
271 {
272 10 XPSNRContext *const s = ctx->priv;
273 10 const uint32_t w = s->plane_width [0]; /* luma image width in pixels */
274 10 const uint32_t h = s->plane_height[0];/* luma image height in pixels */
275 10 const double r = (double)(w * h) / (3840.0 * 2160.0); /* UHD ratio */
276 10 const uint32_t b = FFMAX(0, 4 * (int32_t) (32.0 * sqrt(r) +
277 0.5)); /* block size, integer multiple of 4 for SIMD */
278 10 const uint32_t w_blk = (w + b - 1) / b; /* luma width in units of blocks */
279
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 10 times.
10 const double avg_act = sqrt(16.0 * (double) (1 << (2 * s->depth - 9)) / sqrt(FFMAX(0.00001,
280 r))); /* the sqrt(a_pic) */
281
1/2
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
10 const int *stride_org = (s->bpp == 1 ? s->plane_width : s->line_sizes);
282 10 uint32_t x, y, idx_blk = 0; /* the "16.0" above is due to fixed-point code */
283 10 double *const sse_luma = s->sse_luma;
284 10 double *const weights = s->weights;
285 int c;
286
287
4/8
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 10 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 10 times.
✗ Branch 5 not taken.
✓ Branch 6 taken 10 times.
✗ Branch 7 not taken.
10 if (!wsse64 || (s->depth < 6) || (s->depth > 16) || (s->num_comps <= 0) ||
288
3/6
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 10 times.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✓ Branch 5 taken 10 times.
10 (s->num_comps > 3) || (w == 0) || (h == 0)) {
289 av_log(ctx, AV_LOG_ERROR, "Error in XPSNR routine: invalid argument(s).\n");
290 return AVERROR(EINVAL);
291 }
292
3/6
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 10 times.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✓ Branch 5 taken 10 times.
10 if (!weights || (b >= 4 && !sse_luma)) {
293 av_log(ctx, AV_LOG_ERROR, "Failed to allocate temporary block memory.\n");
294 return AVERROR(ENOMEM);
295 }
296
297
1/2
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
10 if (b >= 4) {
298 10 const int16_t *p_org = org[0];
299 10 const uint32_t s_org = stride_org[0] / s->bpp;
300 10 const int16_t *p_rec = rec[0];
301 10 const uint32_t s_rec = s->plane_width[0];
302 10 double wsse_luma = 0.0;
303
304
2/2
✓ Branch 0 taken 170 times.
✓ Branch 1 taken 10 times.
180 for (y = 0; y < h; y += b) { /* calculate block SSE and perceptual weights */
305
2/2
✓ Branch 0 taken 10 times.
✓ Branch 1 taken 160 times.
170 const uint32_t block_height = (y + b > h ? h - y : b);
306
307
2/2
✓ Branch 0 taken 4250 times.
✓ Branch 1 taken 170 times.
4420 for (x = 0; x < w; x += b, idx_blk++) {
308
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4250 times.
4250 const uint32_t block_width = (x + b > w ? w - x : b);
309 4250 double ms_act = 1.0, ms_act_prev = 0.0;
310
311 8500 sse_luma[idx_blk] = calc_squared_error_and_weight(s, p_org, s_org,
312 org_m1 /* pixel */,
313 org_m2 /* memory */,
314 p_rec, s_rec,
315 x, y,
316 block_width, block_height,
317 4250 s->depth, s->frame_rate, &ms_act);
318 4250 weights[idx_blk] = 1.0 / sqrt(ms_act);
319
320
1/2
✓ Branch 0 taken 4250 times.
✗ Branch 1 not taken.
4250 if (w * h <= 640 * 480) { /* in-line "min-smoothing" as in paper */
321
2/2
✓ Branch 0 taken 170 times.
✓ Branch 1 taken 4080 times.
4250 if (x == 0) /* first column */
322
2/2
✓ Branch 0 taken 160 times.
✓ Branch 1 taken 10 times.
170 ms_act_prev = (idx_blk > 1 ? weights[idx_blk - 2] : 0);
323 else /* after first column */
324
4/4
✓ Branch 0 taken 3910 times.
✓ Branch 1 taken 170 times.
✓ Branch 2 taken 1462 times.
✓ Branch 3 taken 2448 times.
4080 ms_act_prev = (x > b ? FFMAX(weights[idx_blk - 2], weights[idx_blk]) : weights[idx_blk]);
325
326
2/2
✓ Branch 0 taken 3990 times.
✓ Branch 1 taken 260 times.
4250 if (idx_blk > w_blk) /* after the first row and first column */
327
2/2
✓ Branch 0 taken 1801 times.
✓ Branch 1 taken 2189 times.
3990 ms_act_prev = FFMAX(ms_act_prev, weights[idx_blk - 1 - w_blk]); /* min (L, T) */
328
4/4
✓ Branch 0 taken 4240 times.
✓ Branch 1 taken 10 times.
✓ Branch 2 taken 145 times.
✓ Branch 3 taken 4095 times.
4250 if ((idx_blk > 0) && (weights[idx_blk - 1] > ms_act_prev))
329 145 weights[idx_blk - 1] = ms_act_prev;
330
331
5/6
✓ Branch 0 taken 170 times.
✓ Branch 1 taken 4080 times.
✓ Branch 2 taken 10 times.
✓ Branch 3 taken 160 times.
✓ Branch 4 taken 10 times.
✗ Branch 5 not taken.
4250 if ((x + b >= w) && (y + b >= h) && (idx_blk > w_blk)) { /* last block in picture */
332
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 10 times.
10 ms_act_prev = FFMAX(weights[idx_blk - 1], weights[idx_blk - w_blk]);
333
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 6 times.
10 if (weights[idx_blk] > ms_act_prev)
334 4 weights[idx_blk] = ms_act_prev;
335 }
336 }
337 } /* for x */
338 } /* for y */
339
340
2/2
✓ Branch 0 taken 170 times.
✓ Branch 1 taken 10 times.
180 for (y = idx_blk = 0; y < h; y += b) { /* calculate sum for luma (Y) XPSNR */
341
2/2
✓ Branch 0 taken 4250 times.
✓ Branch 1 taken 170 times.
4420 for (x = 0; x < w; x += b, idx_blk++) {
342 4250 wsse_luma += sse_luma[idx_blk] * weights[idx_blk];
343 }
344 }
345
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 10 times.
10 wsse64[0] = (wsse_luma <= 0.0 ? 0 : (uint64_t) (wsse_luma * avg_act + 0.5));
346 } /* b >= 4 */
347
348
2/2
✓ Branch 0 taken 30 times.
✓ Branch 1 taken 10 times.
40 for (c = 0; c < s->num_comps; c++) { /* finalize WSSE value for each component */
349 30 const int16_t *p_org = org[c];
350 30 const uint32_t s_org = stride_org[c] / s->bpp;
351 30 const int16_t *p_rec = rec[c];
352 30 const uint32_t s_rec = s->plane_width[c];
353 30 const uint32_t w_pln = s->plane_width[c];
354 30 const uint32_t h_pln = s->plane_height[c];
355
356
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 30 times.
30 if (b < 4) /* picture is too small for XPSNR, calculate nonweighted PSNR */
357 wsse64[c] = calc_squared_error (s, p_org, s_org,
358 p_rec, s_rec,
359 w_pln, h_pln);
360
2/2
✓ Branch 0 taken 20 times.
✓ Branch 1 taken 10 times.
30 else if (c > 0) { /* b >= 4 so Y XPSNR has already been calculated above */
361 20 const uint32_t bx = (b * w_pln) / w;
362 20 const uint32_t by = (b * h_pln) / h; /* up to chroma downsampling by 4 */
363 20 double wsse_chroma = 0.0;
364
365
2/2
✓ Branch 0 taken 340 times.
✓ Branch 1 taken 20 times.
360 for (y = idx_blk = 0; y < h_pln; y += by) { /* calc chroma (Cb/Cr) XPSNR */
366
2/2
✓ Branch 0 taken 20 times.
✓ Branch 1 taken 320 times.
340 const uint32_t block_height = (y + by > h_pln ? h_pln - y : by);
367
368
2/2
✓ Branch 0 taken 8500 times.
✓ Branch 1 taken 340 times.
8840 for (x = 0; x < w_pln; x += bx, idx_blk++) {
369
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 8500 times.
8500 const uint32_t block_width = (x + bx > w_pln ? w_pln - x : bx);
370
371 17000 wsse_chroma += (double) calc_squared_error (s, p_org + y * s_org + x, s_org,
372 8500 p_rec + y * s_rec + x, s_rec,
373 8500 block_width, block_height) * weights[idx_blk];
374 }
375 }
376
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 20 times.
20 wsse64[c] = (wsse_chroma <= 0.0 ? 0 : (uint64_t) (wsse_chroma * avg_act + 0.5));
377 }
378 } /* for c */
379
380 10 return 0;
381 }
382
383 30 static void set_meta(AVDictionary **metadata, const char *key, char comp, float d)
384 {
385 char value[128];
386 30 snprintf(value, sizeof(value), "%f", d);
387
1/2
✓ Branch 0 taken 30 times.
✗ Branch 1 not taken.
30 if (comp) {
388 char key2[128];
389 30 snprintf(key2, sizeof(key2), "%s%c", key, comp);
390 30 av_dict_set(metadata, key2, value, 0);
391 } else {
392 av_dict_set(metadata, key, value, 0);
393 }
394 30 }
395
396 10 static int do_xpsnr(FFFrameSync *fs)
397 {
398 10 AVFilterContext *ctx = fs->parent;
399 10 XPSNRContext *const s = ctx->priv;
400 10 const uint32_t w = s->plane_width [0]; /* luma image width in pixels */
401 10 const uint32_t h = s->plane_height[0]; /* luma image height in pixels */
402 10 const uint32_t b = FFMAX(0, 4 * (int32_t) (32.0 * sqrt((double) (w * h) / (3840.0 * 2160.0)) + 0.5)); /* block size */
403 10 const uint32_t w_blk = (w + b - 1) / b; /* luma width in units of blocks */
404 10 const uint32_t h_blk = (h + b - 1) / b; /* luma height in units of blocks */
405 10 AVFrame *master, *ref = NULL;
406 int16_t *porg [3];
407 int16_t *prec [3];
408 10 uint64_t wsse64 [3] = {0, 0, 0};
409 10 double cur_xpsnr[3] = {INFINITY, INFINITY, INFINITY};
410 int c, ret_value, stride_org_bpp;
411 AVDictionary **metadata;
412
413
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 10 times.
10 if ((ret_value = ff_framesync_dualinput_get(fs, &master, &ref)) < 0)
414 return ret_value;
415
2/4
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 10 times.
10 if (ctx->is_disabled || !ref)
416 return ff_filter_frame(ctx->outputs[0], master);
417 10 metadata = &master->metadata;
418
419 /* prepare XPSNR calculations: allocate temporary picture and block memory */
420
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 8 times.
10 if (!s->sse_luma)
421 2 s->sse_luma = av_malloc_array(w_blk * h_blk, sizeof(double));
422
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 8 times.
10 if (!s->weights)
423 2 s->weights = av_malloc_array(w_blk * h_blk, sizeof(double));
424
425
2/2
✓ Branch 0 taken 30 times.
✓ Branch 1 taken 10 times.
40 for (c = 0; c < s->num_comps; c++) /* create temporal org buffer memory */
426 30 s->line_sizes[c] = master->linesize[c];
427
428
1/2
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
10 stride_org_bpp = (s->bpp == 1 ? s->plane_width[0] : s->line_sizes[0] / s->bpp);
429
430
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 8 times.
10 if (!s->buf_org_m1)
431 2 s->buf_org_m1 = av_calloc(s->plane_height[0], stride_org_bpp * sizeof(int16_t));
432
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 8 times.
10 if (!s->buf_org_m2)
433 2 s->buf_org_m2 = av_calloc(s->plane_height[0], stride_org_bpp * sizeof(int16_t));
434
435
1/2
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
10 if (s->bpp == 1) { /* 8 bit */
436
2/2
✓ Branch 0 taken 30 times.
✓ Branch 1 taken 10 times.
40 for (c = 0; c < s->num_comps; c++) { /* allocate org/rec buffer memory */
437 30 const int m = s->line_sizes[c]; /* master stride */
438 30 const int r = ref->linesize[c]; /* ref/c stride */
439 30 const int o = s->plane_width[c]; /* XPSNR stride */
440
441
2/2
✓ Branch 0 taken 6 times.
✓ Branch 1 taken 24 times.
30 if (!s->buf_org[c])
442 6 s->buf_org[c] = av_calloc(s->plane_width[c], s->plane_height[c] * sizeof(int16_t));
443
2/2
✓ Branch 0 taken 6 times.
✓ Branch 1 taken 24 times.
30 if (!s->buf_rec[c])
444 6 s->buf_rec[c] = av_calloc(s->plane_width[c], s->plane_height[c] * sizeof(int16_t));
445
446 30 porg[c] = s->buf_org[c];
447 30 prec[c] = s->buf_rec[c];
448
449
2/2
✓ Branch 0 taken 6000 times.
✓ Branch 1 taken 30 times.
6030 for (int y = 0; y < s->plane_height[c]; y++) {
450
2/2
✓ Branch 0 taken 1500000 times.
✓ Branch 1 taken 6000 times.
1506000 for (int x = 0; x < s->plane_width[c]; x++) {
451 1500000 porg[c][y * o + x] = (int16_t) master->data[c][y * m + x];
452 1500000 prec[c][y * o + x] = (int16_t) ref->data[c][y * r + x];
453 }
454 }
455 }
456 } else { /* 10, 12, 14 bit */
457 for (c = 0; c < s->num_comps; c++) {
458 porg[c] = (int16_t *) master->data[c];
459 prec[c] = (int16_t *) ref->data[c];
460 }
461 }
462
463 /* extended perceptually weighted peak signal-to-noise ratio (XPSNR) value */
464 10 ret_value = get_wsse(ctx, porg, s->buf_org_m1, s->buf_org_m2, prec, wsse64);
465
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 10 times.
10 if ( ret_value < 0 )
466 return ret_value; /* an error here means something went wrong earlier! */
467
468
2/2
✓ Branch 0 taken 30 times.
✓ Branch 1 taken 10 times.
40 for (c = 0; c < s->num_comps; c++) {
469 30 const double sqrt_wsse = sqrt((double) wsse64[c]);
470
471 60 cur_xpsnr[c] = get_avg_xpsnr (sqrt_wsse, INFINITY,
472 30 s->plane_width[c], s->plane_height[c],
473 s->max_error_64, 1 /* single frame */);
474 30 s->sum_wdist[c] += sqrt_wsse;
475 30 s->sum_xpsnr[c] += cur_xpsnr[c];
476
1/4
✗ Branch 0 not taken.
✓ Branch 1 taken 30 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
30 s->and_is_inf[c] &= isinf(cur_xpsnr[c]);
477 }
478 10 s->num_frames_64++;
479
480
2/2
✓ Branch 0 taken 30 times.
✓ Branch 1 taken 10 times.
40 for (int j = 0; j < s->num_comps; j++) {
481
2/2
✓ Branch 0 taken 15 times.
✓ Branch 1 taken 15 times.
30 int c = s->is_rgb ? s->rgba_map[j] : j;
482 30 set_meta(metadata, "lavfi.xpsnr.xpsnr.", s->comps[j], cur_xpsnr[c]);
483 }
484
485
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 10 times.
10 if (s->stats_file) { /* print out frame- and component-wise XPSNR averages */
486 fprintf(s->stats_file, "n: %4"PRId64"", s->num_frames_64);
487
488 for (c = 0; c < s->num_comps; c++)
489 fprintf(s->stats_file, " XPSNR %c: %3.4f", s->comps[c], cur_xpsnr[c]);
490 fprintf(s->stats_file, "\n");
491 }
492
493 10 return ff_filter_frame(ctx->outputs[0], master);
494 }
495
496 4 static av_cold int init(AVFilterContext *ctx)
497 {
498 4 XPSNRContext *const s = ctx->priv;
499 int c;
500
501
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
4 if (s->stats_file_str) {
502 if (!strcmp(s->stats_file_str, "-")) /* no stats file, so use stdout */
503 s->stats_file = stdout;
504 else {
505 s->stats_file = avpriv_fopen_utf8(s->stats_file_str, "w");
506
507 if (!s->stats_file) {
508 const int err = AVERROR(errno);
509 char buf[128];
510
511 av_strerror(err, buf, sizeof(buf));
512 av_log(ctx, AV_LOG_ERROR, "Could not open statistics file %s: %s\n", s->stats_file_str, buf);
513 return err;
514 }
515 }
516 }
517
518 4 s->sse_luma = NULL;
519 4 s->weights = NULL;
520
521
2/2
✓ Branch 0 taken 12 times.
✓ Branch 1 taken 4 times.
16 for (c = 0; c < 3; c++) { /* initialize XPSNR data of each color component */
522 12 s->buf_org [c] = NULL;
523 12 s->buf_rec [c] = NULL;
524 12 s->sum_wdist [c] = 0.0;
525 12 s->sum_xpsnr [c] = 0.0;
526 12 s->and_is_inf[c] = 1;
527 }
528
529 4 s->fs.on_event = do_xpsnr;
530
531 4 return 0;
532 }
533
534 static const enum AVPixelFormat xpsnr_formats[] = {
535 AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY9, AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12, AV_PIX_FMT_GRAY14, AV_PIX_FMT_GRAY16,
536 #define PF_NOALPHA(suf) AV_PIX_FMT_YUV420##suf, AV_PIX_FMT_YUV422##suf, AV_PIX_FMT_YUV444##suf
537 #define PF_ALPHA(suf) AV_PIX_FMT_YUVA420##suf, AV_PIX_FMT_YUVA422##suf, AV_PIX_FMT_YUVA444##suf
538 #define PF(suf) PF_NOALPHA(suf), PF_ALPHA(suf)
539 PF(P), PF(P9), PF(P10), PF_NOALPHA(P12), PF_NOALPHA(P14), PF(P16),
540 AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
541 AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P,
542 AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_YUVJ444P,
543 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
544 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
545 AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16,
546 AV_PIX_FMT_NONE
547 };
548
549 2 static int config_input_ref(AVFilterLink *inlink)
550 {
551 2 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
552 2 AVFilterContext *ctx = inlink->dst;
553 2 XPSNRContext *const s = ctx->priv;
554 2 FilterLink *il = ff_filter_link(inlink);
555
556
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
2 if ((ctx->inputs[0]->w != ctx->inputs[1]->w) ||
557
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
2 (ctx->inputs[0]->h != ctx->inputs[1]->h)) {
558 av_log(ctx, AV_LOG_ERROR, "Width and height of the input videos must match.\n");
559 return AVERROR(EINVAL);
560 }
561
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
2 if (ctx->inputs[0]->format != ctx->inputs[1]->format) {
562 av_log(ctx, AV_LOG_ERROR, "The input videos must be of the same pixel format.\n");
563 return AVERROR(EINVAL);
564 }
565
566
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
2 s->bpp = (desc->comp[0].depth <= 8 ? 1 : 2);
567 2 s->depth = desc->comp[0].depth;
568 2 s->max_error_64 = (1 << s->depth) - 1; /* conventional limit */
569 2 s->max_error_64 *= s->max_error_64;
570
571 2 s->frame_rate = il->frame_rate.num / il->frame_rate.den;
572
573 2 s->num_comps = (desc->nb_components > 3 ? 3 : desc->nb_components);
574
575 2 s->is_rgb = (ff_fill_rgba_map(s->rgba_map, inlink->format) >= 0);
576
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
2 s->comps[0] = (s->is_rgb ? 'r' : 'y');
577
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
2 s->comps[1] = (s->is_rgb ? 'g' : 'u');
578
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
2 s->comps[2] = (s->is_rgb ? 'b' : 'v');
579 2 s->comps[3] = 'a';
580
581 2 s->plane_width [1] = s->plane_width [2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
582 2 s->plane_width [0] = s->plane_width [3] = inlink->w;
583 2 s->plane_height[1] = s->plane_height[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
584 2 s->plane_height[0] = s->plane_height[3] = inlink->h;
585
586 /* XPSNR always operates with 16-bit internal precision */
587 2 ff_psnr_init(&s->pdsp, 15);
588 2 s->dsp.highds_func = highds; /* initialize filtering methods */
589 2 s->dsp.diff1st_func = diff1st;
590 2 s->dsp.diff2nd_func = diff2nd;
591
592 2 return 0;
593 }
594
595 2 static int config_output(AVFilterLink *outlink)
596 {
597 2 AVFilterContext *ctx = outlink->src;
598 2 XPSNRContext *s = ctx->priv;
599 2 AVFilterLink *mainlink = ctx->inputs[0];
600 2 FilterLink *il = ff_filter_link(mainlink);
601 2 FilterLink *ol = ff_filter_link(outlink);
602 int ret;
603
604
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
2 if ((ret = ff_framesync_init_dualinput(&s->fs, ctx)) < 0)
605 return ret;
606
607 2 outlink->w = mainlink->w;
608 2 outlink->h = mainlink->h;
609 2 outlink->time_base = mainlink->time_base;
610 2 outlink->sample_aspect_ratio = mainlink->sample_aspect_ratio;
611 2 ol->frame_rate = il->frame_rate;
612
613
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
2 if ((ret = ff_framesync_configure(&s->fs)) < 0)
614 return ret;
615
616 2 outlink->time_base = s->fs.time_base;
617
618
2/4
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✓ Branch 4 taken 2 times.
4 if (av_cmp_q(mainlink->time_base, outlink->time_base) ||
619 2 av_cmp_q(ctx->inputs[1]->time_base, outlink->time_base))
620 av_log(ctx, AV_LOG_WARNING, "not matching timebases found between first input: %d/%d and second input %d/%d, results may be incorrect!\n",
621 mainlink->time_base.num, mainlink->time_base.den,
622 ctx->inputs[1]->time_base.num, ctx->inputs[1]->time_base.den);
623
624 2 return 0;
625 }
626
627 30 static int activate(AVFilterContext *ctx)
628 {
629 30 XPSNRContext *s = ctx->priv;
630
631 30 return ff_framesync_activate(&s->fs);
632 }
633
634 4 static av_cold void uninit(AVFilterContext *ctx)
635 {
636 4 XPSNRContext *const s = ctx->priv;
637 int c;
638
639
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
4 if (s->num_frames_64 > 0) { /* print out overall component-wise mean XPSNR */
640 2 const double xpsnr_luma = get_avg_xpsnr(s->sum_wdist[0], s->sum_xpsnr[0],
641 2 s->plane_width[0], s->plane_height[0],
642 s->max_error_64, s->num_frames_64);
643 2 double xpsnr_min = xpsnr_luma;
644
645 /* luma */
646 2 av_log(ctx, AV_LOG_INFO, "XPSNR %c: %3.4f", s->comps[0], xpsnr_luma);
647
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
2 if (s->stats_file) {
648 fprintf(s->stats_file, "\nXPSNR average, %"PRId64" frames", s->num_frames_64);
649 fprintf(s->stats_file, " %c: %3.4f", s->comps[0], xpsnr_luma);
650 }
651 /* chroma */
652
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 2 times.
6 for (c = 1; c < s->num_comps; c++) {
653 4 const double xpsnr_chroma = get_avg_xpsnr(s->sum_wdist[c], s->sum_xpsnr[c],
654 4 s->plane_width[c], s->plane_height[c],
655 s->max_error_64, s->num_frames_64);
656
2/2
✓ Branch 0 taken 3 times.
✓ Branch 1 taken 1 times.
4 if (xpsnr_min > xpsnr_chroma)
657 3 xpsnr_min = xpsnr_chroma;
658
659 4 av_log(ctx, AV_LOG_INFO, " %c: %3.4f", s->comps[c], xpsnr_chroma);
660
1/4
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
4 if (s->stats_file && s->stats_file != stdout)
661 fprintf(s->stats_file, " %c: %3.4f", s->comps[c], xpsnr_chroma);
662 }
663 /* print out line break, and minimum XPSNR across the color components */
664
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
2 if (s->num_comps > 1) {
665 2 av_log(ctx, AV_LOG_INFO, " (minimum: %3.4f)\n", xpsnr_min);
666
1/4
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
2 if (s->stats_file && s->stats_file != stdout)
667 fprintf(s->stats_file, " (minimum: %3.4f)\n", xpsnr_min);
668 } else {
669 av_log(ctx, AV_LOG_INFO, "\n");
670 if (s->stats_file && s->stats_file != stdout)
671 fprintf(s->stats_file, "\n");
672 }
673 }
674
675 4 ff_framesync_uninit(&s->fs); /* free temporary picture or block buf memory */
676
677
1/4
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
4 if (s->stats_file && s->stats_file != stdout)
678 fclose(s->stats_file);
679
680 4 av_freep(&s->sse_luma);
681 4 av_freep(&s->weights );
682
683 4 av_freep(&s->buf_org_m1);
684 4 av_freep(&s->buf_org_m2);
685
686
2/2
✓ Branch 0 taken 6 times.
✓ Branch 1 taken 4 times.
10 for (c = 0; c < s->num_comps; c++) {
687 6 av_freep(&s->buf_org[c]);
688 6 av_freep(&s->buf_rec[c]);
689 }
690 4 }
691
692 static const AVFilterPad xpsnr_inputs[] = {
693 {
694 .name = "main",
695 .type = AVMEDIA_TYPE_VIDEO,
696 }, {
697 .name = "reference",
698 .type = AVMEDIA_TYPE_VIDEO,
699 .config_props = config_input_ref,
700 }
701 };
702
703 static const AVFilterPad xpsnr_outputs[] = {
704 {
705 .name = "default",
706 .type = AVMEDIA_TYPE_VIDEO,
707 .config_props = config_output,
708 }
709 };
710
711 const FFFilter ff_vf_xpsnr = {
712 .p.name = "xpsnr",
713 .p.description = NULL_IF_CONFIG_SMALL("Calculate the extended perceptually weighted peak signal-to-noise ratio (XPSNR) between two video streams."),
714 .p.priv_class = &xpsnr_class,
715 .p.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_METADATA_ONLY,
716 .preinit = xpsnr_framesync_preinit,
717 .init = init,
718 .uninit = uninit,
719 .activate = activate,
720 .priv_size = sizeof(XPSNRContext),
721 FILTER_INPUTS (xpsnr_inputs),
722 FILTER_OUTPUTS(xpsnr_outputs),
723 FILTER_PIXFMTS_ARRAY(xpsnr_formats),
724 };
725