1 /*
2 * Copyright (c) 2019 Vladimir Panteleev
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 #include <float.h>
22
23 #include "libavutil/imgutils.h"
24 #include "libavutil/opt.h"
25 #include "libavutil/pixdesc.h"
26 #include "avfilter.h"
27
28 #include "formats.h"
29 #include "internal.h"
30 #include "video.h"
31
32 #define MAX_FRAMES 240
33 #define GRID_SIZE 8
34 #define NUM_CHANNELS 3
35
36 typedef struct PhotosensitivityFrame {
37 uint8_t grid[GRID_SIZE][GRID_SIZE][4];
38 } PhotosensitivityFrame;
39
40 typedef struct PhotosensitivityContext {
41 const AVClass *class;
42
43 int nb_frames;
44 int skip;
45 float threshold_multiplier;
46 int bypass;
47
48 int badness_threshold;
49
50 /* Circular buffer */
51 int history[MAX_FRAMES];
52 int history_pos;
53
54 PhotosensitivityFrame last_frame_e;
55 AVFrame *last_frame_av;
56 } PhotosensitivityContext;
57
58 #define OFFSET(x) offsetof(PhotosensitivityContext, x)
59 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
60
61 static const AVOption photosensitivity_options[] = {
62 { "frames", "set how many frames to use", OFFSET(nb_frames), AV_OPT_TYPE_INT, {.i64=30}, 2, MAX_FRAMES, FLAGS },
63 { "f", "set how many frames to use", OFFSET(nb_frames), AV_OPT_TYPE_INT, {.i64=30}, 2, MAX_FRAMES, FLAGS },
64 { "threshold", "set detection threshold factor (lower is stricter)", OFFSET(threshold_multiplier), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0.1, FLT_MAX, FLAGS },
65 { "t", "set detection threshold factor (lower is stricter)", OFFSET(threshold_multiplier), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0.1, FLT_MAX, FLAGS },
66 { "skip", "set pixels to skip when sampling frames", OFFSET(skip), AV_OPT_TYPE_INT, {.i64=1}, 1, 1024, FLAGS },
67 { "bypass", "leave frames unchanged", OFFSET(bypass), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS },
68 { NULL }
69 };
70
71 AVFILTER_DEFINE_CLASS(photosensitivity);
72
73 typedef struct ThreadData_convert_frame
74 {
75 AVFrame *in;
76 PhotosensitivityFrame *out;
77 int skip;
78 } ThreadData_convert_frame;
79
80 #define NUM_CELLS (GRID_SIZE * GRID_SIZE)
81
convert_frame_partial(AVFilterContext * ctx,void * arg,int jobnr,int nb_jobs)82 static int convert_frame_partial(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
83 {
84 int cell, gx, gy, x0, x1, y0, y1, x, y, c, area;
85 int sum[NUM_CHANNELS];
86 const uint8_t *p;
87
88 ThreadData_convert_frame *td = arg;
89
90 const int slice_start = (NUM_CELLS * jobnr) / nb_jobs;
91 const int slice_end = (NUM_CELLS * (jobnr+1)) / nb_jobs;
92
93 int width = td->in->width, height = td->in->height, linesize = td->in->linesize[0], skip = td->skip;
94 const uint8_t *data = td->in->data[0];
95
96 for (cell = slice_start; cell < slice_end; cell++) {
97 gx = cell % GRID_SIZE;
98 gy = cell / GRID_SIZE;
99
100 x0 = width * gx / GRID_SIZE;
101 x1 = width * (gx+1) / GRID_SIZE;
102 y0 = height * gy / GRID_SIZE;
103 y1 = height * (gy+1) / GRID_SIZE;
104
105 for (c = 0; c < NUM_CHANNELS; c++) {
106 sum[c] = 0;
107 }
108 for (y = y0; y < y1; y += skip) {
109 p = data + y * linesize + x0 * NUM_CHANNELS;
110 for (x = x0; x < x1; x += skip) {
111 //av_log(NULL, AV_LOG_VERBOSE, "%d %d %d : (%d,%d) (%d,%d) -> %d,%d | *%d\n", c, gx, gy, x0, y0, x1, y1, x, y, (int)row);
112 sum[0] += p[0];
113 sum[1] += p[1];
114 sum[2] += p[2];
115 p += NUM_CHANNELS * skip;
116 // TODO: variable size
117 }
118 }
119
120 area = ((x1 - x0 + skip - 1) / skip) * ((y1 - y0 + skip - 1) / skip);
121 for (c = 0; c < NUM_CHANNELS; c++) {
122 if (area)
123 sum[c] /= area;
124 td->out->grid[gy][gx][c] = sum[c];
125 }
126 }
127 return 0;
128 }
129
convert_frame(AVFilterContext * ctx,AVFrame * in,PhotosensitivityFrame * out,int skip)130 static void convert_frame(AVFilterContext *ctx, AVFrame *in, PhotosensitivityFrame *out, int skip)
131 {
132 ThreadData_convert_frame td;
133 td.in = in;
134 td.out = out;
135 td.skip = skip;
136 ff_filter_execute(ctx, convert_frame_partial, &td, NULL,
137 FFMIN(NUM_CELLS, ff_filter_get_nb_threads(ctx)));
138 }
139
140 typedef struct ThreadData_blend_frame
141 {
142 AVFrame *target;
143 AVFrame *source;
144 uint16_t s_mul;
145 } ThreadData_blend_frame;
146
blend_frame_partial(AVFilterContext * ctx,void * arg,int jobnr,int nb_jobs)147 static int blend_frame_partial(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
148 {
149 int x, y;
150 uint8_t *t, *s;
151
152 ThreadData_blend_frame *td = arg;
153 const uint16_t s_mul = td->s_mul;
154 const uint16_t t_mul = 0x100 - s_mul;
155 const int slice_start = (td->target->height * jobnr) / nb_jobs;
156 const int slice_end = (td->target->height * (jobnr+1)) / nb_jobs;
157 const int linesize = td->target->linesize[0];
158
159 for (y = slice_start; y < slice_end; y++) {
160 t = td->target->data[0] + y * td->target->linesize[0];
161 s = td->source->data[0] + y * td->source->linesize[0];
162 for (x = 0; x < linesize; x++) {
163 *t = (*t * t_mul + *s * s_mul) >> 8;
164 t++; s++;
165 }
166 }
167 return 0;
168 }
169
blend_frame(AVFilterContext * ctx,AVFrame * target,AVFrame * source,float factor)170 static void blend_frame(AVFilterContext *ctx, AVFrame *target, AVFrame *source, float factor)
171 {
172 ThreadData_blend_frame td;
173 td.target = target;
174 td.source = source;
175 td.s_mul = (uint16_t)(factor * 0x100);
176 ff_filter_execute(ctx, blend_frame_partial, &td, NULL,
177 FFMIN(ctx->outputs[0]->h, ff_filter_get_nb_threads(ctx)));
178 }
179
get_badness(PhotosensitivityFrame * a,PhotosensitivityFrame * b)180 static int get_badness(PhotosensitivityFrame *a, PhotosensitivityFrame *b)
181 {
182 int badness, x, y, c;
183 badness = 0;
184 for (c = 0; c < NUM_CHANNELS; c++) {
185 for (y = 0; y < GRID_SIZE; y++) {
186 for (x = 0; x < GRID_SIZE; x++) {
187 badness += abs((int)a->grid[y][x][c] - (int)b->grid[y][x][c]);
188 //av_log(NULL, AV_LOG_VERBOSE, "%d - %d -> %d \n", a->grid[y][x], b->grid[y][x], badness);
189 //av_log(NULL, AV_LOG_VERBOSE, "%d -> %d \n", abs((int)a->grid[y][x] - (int)b->grid[y][x]), badness);
190 }
191 }
192 }
193 return badness;
194 }
195
config_input(AVFilterLink * inlink)196 static int config_input(AVFilterLink *inlink)
197 {
198 /* const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format); */
199 AVFilterContext *ctx = inlink->dst;
200 PhotosensitivityContext *s = ctx->priv;
201
202 s->badness_threshold = (int)(GRID_SIZE * GRID_SIZE * 4 * 256 * s->nb_frames * s->threshold_multiplier / 128);
203
204 return 0;
205 }
206
filter_frame(AVFilterLink * inlink,AVFrame * in)207 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
208 {
209 int this_badness, current_badness, fixed_badness, new_badness, i, res;
210 PhotosensitivityFrame ef;
211 AVFrame *src, *out;
212 int free_in = 0;
213 float factor;
214 AVDictionary **metadata;
215
216 AVFilterContext *ctx = inlink->dst;
217 AVFilterLink *outlink = ctx->outputs[0];
218 PhotosensitivityContext *s = ctx->priv;
219
220 /* weighted moving average */
221 current_badness = 0;
222 for (i = 1; i < s->nb_frames; i++)
223 current_badness += i * s->history[(s->history_pos + i) % s->nb_frames];
224 current_badness /= s->nb_frames;
225
226 convert_frame(ctx, in, &ef, s->skip);
227 this_badness = get_badness(&ef, &s->last_frame_e);
228 new_badness = current_badness + this_badness;
229 av_log(s, AV_LOG_VERBOSE, "badness: %6d -> %6d / %6d (%3d%% - %s)\n",
230 current_badness, new_badness, s->badness_threshold,
231 100 * new_badness / s->badness_threshold, new_badness < s->badness_threshold ? "OK" : "EXCEEDED");
232
233 fixed_badness = new_badness;
234 if (new_badness < s->badness_threshold || !s->last_frame_av || s->bypass) {
235 factor = 1; /* for metadata */
236 av_frame_free(&s->last_frame_av);
237 s->last_frame_av = src = in;
238 s->last_frame_e = ef;
239 s->history[s->history_pos] = this_badness;
240 } else {
241 factor = (float)(s->badness_threshold - current_badness) / (new_badness - current_badness);
242 if (factor <= 0) {
243 /* just duplicate the frame */
244 s->history[s->history_pos] = 0; /* frame was duplicated, thus, delta is zero */
245 } else {
246 res = av_frame_make_writable(s->last_frame_av);
247 if (res) {
248 av_frame_free(&in);
249 return res;
250 }
251 blend_frame(ctx, s->last_frame_av, in, factor);
252
253 convert_frame(ctx, s->last_frame_av, &ef, s->skip);
254 this_badness = get_badness(&ef, &s->last_frame_e);
255 fixed_badness = current_badness + this_badness;
256 av_log(s, AV_LOG_VERBOSE, " fixed: %6d -> %6d / %6d (%3d%%) factor=%5.3f\n",
257 current_badness, fixed_badness, s->badness_threshold,
258 100 * new_badness / s->badness_threshold, factor);
259 s->last_frame_e = ef;
260 s->history[s->history_pos] = this_badness;
261 }
262 src = s->last_frame_av;
263 free_in = 1;
264 }
265 s->history_pos = (s->history_pos + 1) % s->nb_frames;
266
267 out = ff_get_video_buffer(outlink, in->width, in->height);
268 if (!out) {
269 if (free_in == 1)
270 av_frame_free(&in);
271 return AVERROR(ENOMEM);
272 }
273 av_frame_copy_props(out, in);
274 metadata = &out->metadata;
275 if (metadata) {
276 char value[128];
277
278 snprintf(value, sizeof(value), "%f", (float)new_badness / s->badness_threshold);
279 av_dict_set(metadata, "lavfi.photosensitivity.badness", value, 0);
280
281 snprintf(value, sizeof(value), "%f", (float)fixed_badness / s->badness_threshold);
282 av_dict_set(metadata, "lavfi.photosensitivity.fixed-badness", value, 0);
283
284 snprintf(value, sizeof(value), "%f", (float)this_badness / s->badness_threshold);
285 av_dict_set(metadata, "lavfi.photosensitivity.frame-badness", value, 0);
286
287 snprintf(value, sizeof(value), "%f", factor);
288 av_dict_set(metadata, "lavfi.photosensitivity.factor", value, 0);
289 }
290 av_frame_copy(out, src);
291 if (free_in == 1)
292 av_frame_free(&in);
293 return ff_filter_frame(outlink, out);
294 }
295
uninit(AVFilterContext * ctx)296 static av_cold void uninit(AVFilterContext *ctx)
297 {
298 PhotosensitivityContext *s = ctx->priv;
299
300 av_frame_free(&s->last_frame_av);
301 }
302
303 static const AVFilterPad inputs[] = {
304 {
305 .name = "default",
306 .type = AVMEDIA_TYPE_VIDEO,
307 .filter_frame = filter_frame,
308 .config_props = config_input,
309 },
310 };
311
312 static const AVFilterPad outputs[] = {
313 {
314 .name = "default",
315 .type = AVMEDIA_TYPE_VIDEO,
316 },
317 };
318
319 const AVFilter ff_vf_photosensitivity = {
320 .name = "photosensitivity",
321 .description = NULL_IF_CONFIG_SMALL("Filter out photosensitive epilepsy seizure-inducing flashes."),
322 .priv_size = sizeof(PhotosensitivityContext),
323 .priv_class = &photosensitivity_class,
324 .uninit = uninit,
325 FILTER_INPUTS(inputs),
326 FILTER_OUTPUTS(outputs),
327 FILTER_PIXFMTS(AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24),
328 };
329