History

GSoC: Add guided filter

Add examples on how to use this filter, and improve the code style. Implement the slice-level parallelism for guided filter. Add the basic version of guided filter.

Xuewei Meng committed 5 years ago

Commit f8d910e1 parent 8649f5d4 files · +469 −0Browse files

Showing the changes to libavfilter (3 of 4 files). Show all

Unified · Split

3 files
  1. libavfilter/Makefile +1 −0
  2. libavfilter/allfilters.c +1 −0
  3. libavfilter/vf_guided.c +429 −0
libavfilter/Makefile+1−0
@@ -286,6 +286,7 @@ OBJS-$(CONFIG_GEQ_FILTER) += vf_geq.o · 285 unmodified lines
286OBJS-$(CONFIG_GRADFUN_FILTER) += vf_gradfun.o286OBJS-$(CONFIG_GRADFUN_FILTER) += vf_gradfun.o
287OBJS-$(CONFIG_GRAPHMONITOR_FILTER) += f_graphmonitor.o287OBJS-$(CONFIG_GRAPHMONITOR_FILTER) += f_graphmonitor.o
288OBJS-$(CONFIG_GREYEDGE_FILTER) += vf_colorconstancy.o288OBJS-$(CONFIG_GREYEDGE_FILTER) += vf_colorconstancy.o
289OBJS-$(CONFIG_GUIDED_FILTER) += vf_guided.o
289OBJS-$(CONFIG_HALDCLUT_FILTER) += vf_lut3d.o framesync.o290OBJS-$(CONFIG_HALDCLUT_FILTER) += vf_lut3d.o framesync.o
290OBJS-$(CONFIG_HFLIP_FILTER) += vf_hflip.o291OBJS-$(CONFIG_HFLIP_FILTER) += vf_hflip.o
291OBJS-$(CONFIG_HISTEQ_FILTER) += vf_histeq.o292OBJS-$(CONFIG_HISTEQ_FILTER) += vf_histeq.o
libavfilter/allfilters.c+1−0
@@ -271,6 +271,7 @@ extern const AVFilter ff_vf_geq; · 270 unmodified lines
271extern const AVFilter ff_vf_gradfun;271extern const AVFilter ff_vf_gradfun;
272extern const AVFilter ff_vf_graphmonitor;272extern const AVFilter ff_vf_graphmonitor;
273extern const AVFilter ff_vf_greyedge;273extern const AVFilter ff_vf_greyedge;
274extern const AVFilter ff_vf_guided;
274extern const AVFilter ff_vf_haldclut;275extern const AVFilter ff_vf_haldclut;
275extern const AVFilter ff_vf_hflip;276extern const AVFilter ff_vf_hflip;
276extern const AVFilter ff_vf_histeq;277extern const AVFilter ff_vf_histeq;
libavfilter/vf_guided.c+429−0
@@ -0,0 +1,429 @@
1/*
2 * Copyright (c) 2021 Xuewei Meng
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 "libavutil/imgutils.h"
22#include "libavutil/opt.h"
23#include "libavutil/pixdesc.h"
24#include "avfilter.h"
25#include "formats.h"
26#include "framesync.h"
27#include "internal.h"
28#include "video.h"
29
30typedef struct GuidedContext {
31 const AVClass *class;
32 FFFrameSync fs;
33
34 int radius;
35 float eps;
36
37 int planes;
38
39 int width;
40 int height;
41
42 int nb_planes;
43 int depth;
44 int planewidth[4];
45 int planeheight[4];
46
47 int (*box_slice)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
48} GuidedContext;
49
50#define OFFSET(x) offsetof(GuidedContext, x)
51#define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
52
53static const AVOption guided_options[] = {
54 { "radius", "set the box radius", OFFSET(radius), AV_OPT_TYPE_INT, {.i64=3 }, 1, 20, FLAGS },
55 { "eps", "set the regularization parameter (with square)", OFFSET(eps), AV_OPT_TYPE_FLOAT, {.dbl=0.01 }, 0.0, 1, FLAGS },
56 { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=1 }, 0, 0xF, FLAGS },
57 { NULL }
58};
59
60AVFILTER_DEFINE_CLASS(guided);
61
62typedef struct ThreadData {
63 int width;
64 int height;
65 float *src;
66 float *dst;
67 int srcStride;
68 int dstStride;
69} ThreadData;
70
71static int box_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
72{
73 GuidedContext *s = ctx->priv;
74 ThreadData *t = arg;
75
76 const int width = t->width;
77 const int height = t->height;
78 const int src_stride = t->srcStride;
79 const int dst_stride = t->dstStride;
80 const int slice_start = (height * jobnr) / nb_jobs;
81 const int slice_end = (height * (jobnr + 1)) / nb_jobs;
82 const int radius = s->radius;
83 const float *src = t->src;
84 float *dst = t->dst;
85
86 int w;
87 int numPix;
88 w = (radius << 1) + 1;
89 numPix = w * w;
90 for (int i = slice_start;i < slice_end;i++) {
91 for (int j = 0;j < width;j++) {
92 float temp = 0.0;
93 for (int row = -radius;row <= radius;row++) {
94 for (int col = -radius;col <= radius;col++) {
95 int x = i + row;
96 int y = j + col;
97 x = (x < 0) ? 0 : (x >= height ? height - 1 : x);
98 y = (y < 0) ? 0 : (y >= width ? width - 1 : y);
99 temp += src[x * src_stride + y];
100 }
101 }
102 dst[i * dst_stride + j] = temp / numPix;
103 }
104 }
105 return 0;
106}
107
108static int query_formats(AVFilterContext *ctx)
109{
110 static const enum AVPixelFormat pix_fmts[] = {
111 AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV440P,
112 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P,
113 AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUV420P,
114 AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ420P,
115 AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
116 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
117 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
118 AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV440P12,
119 AV_PIX_FMT_YUV420P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV444P14,
120 AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
121 AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA444P9,
122 AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_YUVA444P10,
123 AV_PIX_FMT_YUVA420P16, AV_PIX_FMT_YUVA422P16, AV_PIX_FMT_YUVA444P16,
124 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
125 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
126 AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16,
127 AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY9, AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12, AV_PIX_FMT_GRAY14, AV_PIX_FMT_GRAY16,
128 AV_PIX_FMT_NONE
129 };
130
131 return ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
132}
133
134static int config_input(AVFilterLink *inlink)
135{
136 AVFilterContext *ctx = inlink->dst;
137 GuidedContext *s = ctx->priv;
138 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
139
140 if (ctx->inputs[0]->w != ctx->inputs[1]->w ||
141 ctx->inputs[0]->h != ctx->inputs[1]->h) {
142 av_log(ctx, AV_LOG_ERROR, "Width and height of input videos must be same.\n");
143 return AVERROR(EINVAL);
144 }
145 if (ctx->inputs[0]->format != ctx->inputs[1]->format) {
146 av_log(ctx, AV_LOG_ERROR, "Inputs must be of same pixel format.\n");
147 return AVERROR(EINVAL);
148 }
149
150
151 s->depth = desc->comp[0].depth;
152 s->width = ctx->inputs[0]->w;
153 s->height = ctx->inputs[0]->h;
154
155 s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
156 s->planewidth[0] = s->planewidth[3] = inlink->w;
157 s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
158 s->planeheight[0] = s->planeheight[3] = inlink->h;
159
160 s->nb_planes = av_pix_fmt_count_planes(inlink->format);
161 s->box_slice = box_slice;
162 return 0;
163}
164
165#define GUIDED(type, name) \
166static int guided_##name(AVFilterContext *ctx, GuidedContext *s, \
167 const uint8_t *ssrc, const uint8_t *ssrcRef, \
168 uint8_t *ddst, int radius, float eps, int width, int height, \
169 int src_stride, int src_ref_stride, int dst_stride, \
170 float maxval) \
171{ \
172 int ret = 0; \
173 type *dst = (type *)ddst; \
174 const type *src = (const type *)ssrc; \
175 const type *srcRef = (const type *)ssrcRef; \
176 \
177 ThreadData t; \
178 const int nb_threads = ff_filter_get_nb_threads(ctx); \
179 float *I; \
180 float *II; \
181 float *P; \
182 float *IP; \
183 float *meanI; \
184 float *meanII; \
185 float *meanP; \
186 float *meanIP; \
187 float *A; \
188 float *B; \
189 float *meanA; \
190 float *meanB; \
191 \
192 I = av_calloc(width * height, sizeof(float)); \
193 II = av_calloc(width * height, sizeof(float)); \
194 P = av_calloc(width * height, sizeof(float)); \
195 IP = av_calloc(width * height, sizeof(float)); \
196 meanI = av_calloc(width * height, sizeof(float)); \
197 meanII = av_calloc(width * height, sizeof(float)); \
198 meanP = av_calloc(width * height, sizeof(float)); \
199 meanIP = av_calloc(width * height, sizeof(float)); \
200 \
201 A = av_calloc(width * height, sizeof(float)); \
202 B = av_calloc(width * height, sizeof(float)); \
203 meanA = av_calloc(width * height, sizeof(float)); \
204 meanB = av_calloc(width * height, sizeof(float)); \
205 \
206 if (!I || !II || !P || !IP || !meanI || !meanII || !meanP || \
207 !meanIP || !A || !B || !meanA || !meanB){ \
208 ret = AVERROR(ENOMEM); \
209 goto end; \
210 } \
211 for (int i = 0;i < height;i++) { \
212 for (int j = 0;j < width;j++) { \
213 int x = i * width + j; \
214 I[x] = src[i * src_stride + j] / maxval; \
215 II[x] = I[x] * I[x]; \
216 P[x] = srcRef[i * src_ref_stride + j] / maxval; \
217 IP[x] = I[x] * P[x]; \
218 } \
219 } \
220 \
221 t.width = width; \
222 t.height = height; \
223 t.srcStride = width; \
224 t.dstStride = width; \
225 t.src = I; \
226 t.dst = meanI; \
227 ctx->internal->execute(ctx, s->box_slice, &t, NULL, FFMIN(height, nb_threads)); \
228 t.src = II; \
229 t.dst = meanII; \
230 ctx->internal->execute(ctx, s->box_slice, &t, NULL, FFMIN(height, nb_threads)); \
231 t.src = P; \
232 t.dst = meanP; \
233 ctx->internal->execute(ctx, s->box_slice, &t, NULL, FFMIN(height, nb_threads)); \
234 t.src = IP; \
235 t.dst = meanIP; \
236 ctx->internal->execute(ctx, s->box_slice, &t, NULL, FFMIN(height, nb_threads)); \
237 \
238 for (int i = 0;i < height;i++) { \
239 for (int j = 0;j < width;j++) { \
240 int x = i * width + j; \
241 float varI = meanII[x] - (meanI[x] * meanI[x]); \
242 float covIP = meanIP[x] - (meanI[x] * meanP[x]); \
243 A[x] = covIP / (varI + eps); \
244 B[x] = meanP[x] - A[x] * meanI[x]; \
245 } \
246 } \
247 \
248 t.src = A; \
249 t.dst = meanA; \
250 ctx->internal->execute(ctx, s->box_slice, &t, NULL, FFMIN(height, nb_threads)); \
251 t.src = B; \
252 t.dst = meanB; \
253 ctx->internal->execute(ctx, s->box_slice, &t, NULL, FFMIN(height, nb_threads)); \
254 \
255 for (int i = 0;i < height;i++) { \
256 for (int j = 0;j < width;j++) { \
257 int x = i * width + j; \
258 dst[i * dst_stride + j] = meanA[x] * src[i * src_stride + j] + \
259 meanB[x] * maxval; \
260 } \
261 } \
262end: \
263 av_freep(&I); \
264 av_freep(&II); \
265 av_freep(&P); \
266 av_freep(&IP); \
267 av_freep(&meanI); \
268 av_freep(&meanII); \
269 av_freep(&meanP); \
270 av_freep(&meanIP); \
271 av_freep(&A); \
272 av_freep(&B); \
273 av_freep(&meanA); \
274 av_freep(&meanB); \
275 return ret; \
276}
277
278GUIDED(uint8_t, byte)
279GUIDED(uint16_t, word)
280
281static int process_frame(FFFrameSync *fs)
282{
283 AVFilterContext *ctx = fs->parent;
284 GuidedContext *s = fs->opaque;
285 AVFilterLink *outlink = ctx->outputs[0];
286 AVFrame *out_frame = NULL, *main_frame = NULL, *ref_frame = NULL;
287 int ret;
288
289 ret = ff_framesync_dualinput_get(fs, &main_frame, &ref_frame);
290 if (ret < 0)
291 return ret;
292
293 out_frame = ff_get_video_buffer(outlink, outlink->w, outlink->h);
294 if (!out_frame) {
295 av_frame_free(&main_frame);
296 av_frame_free(&ref_frame);
297 return AVERROR(ENOMEM);
298 }
299 av_frame_copy_props(out_frame, main_frame);
300
301 if (ctx->is_disabled || !ref_frame) {
302 av_frame_copy_props(ref_frame, main_frame);
303 }
304
305 for (int plane = 0; plane < s->nb_planes; plane++) {
306 if (!(s->planes & (1 << plane))) {
307 av_image_copy_plane(out_frame->data[plane], out_frame->linesize[plane],
308 main_frame->data[plane], main_frame->linesize[plane],
309 s->planewidth[plane] * ((s->depth + 7) / 8), s->planeheight[plane]);
310 continue;
311 }
312 if (s->depth <= 8)
313 guided_byte(ctx, s, main_frame->data[plane], ref_frame->data[plane], out_frame->data[plane], s->radius, s->eps,
314 s->planewidth[plane], s->planeheight[plane],
315 main_frame->linesize[plane], ref_frame->linesize[plane], out_frame->linesize[plane], (1 << s->depth) - 1.f);
316 else
317 guided_word(ctx, s, main_frame->data[plane], ref_frame->data[plane], out_frame->data[plane], s->radius, s->eps,
318 s->planewidth[plane], s->planeheight[plane],
319 main_frame->linesize[plane] / 2, ref_frame->linesize[plane] / 2, out_frame->linesize[plane] / 2, (1 << s->depth) - 1.f);
320 }
321
322 return ff_filter_frame(outlink, out_frame);
323}
324
325
326static int config_output(AVFilterLink *outlink)
327{
328 AVFilterContext *ctx = outlink->src;
329
330 GuidedContext *s = ctx->priv;
331 AVFilterLink *mainlink = ctx->inputs[0];
332 FFFrameSyncIn *in;
333 int ret;
334
335
336 outlink->w = mainlink->w;
337 outlink->h = mainlink->h;
338 outlink->time_base = mainlink->time_base;
339 outlink->sample_aspect_ratio = mainlink->sample_aspect_ratio;
340 outlink->frame_rate = mainlink->frame_rate;
341 if ((ret = ff_framesync_init(&s->fs, ctx, 2)) < 0)
342 return ret;
343
344 outlink->time_base = s->fs.time_base;
345
346 in = s->fs.in;
347 in[0].time_base = mainlink->time_base;
348 in[1].time_base = ctx->inputs[1]->time_base;
349 in[0].sync = 2;
350 in[0].before = EXT_INFINITY;
351 in[0].after = EXT_INFINITY;
352 in[1].sync = 1;
353 in[1].before = EXT_INFINITY;
354 in[1].after = EXT_INFINITY;
355 s->fs.opaque = s;
356 s->fs.on_event = process_frame;
357
358 return ff_framesync_configure(&s->fs);
359}
360
361static int activate(AVFilterContext *ctx)
362{
363 GuidedContext *s = ctx->priv;
364 return ff_framesync_activate(&s->fs);
365}
366
367static av_cold int init(AVFilterContext *ctx)
368{
369 return 0;
370}
371
372static av_cold void uninit(AVFilterContext *ctx)
373{
374 GuidedContext *s = ctx->priv;
375 ff_framesync_uninit(&s->fs);
376 return;
377}
378
379
380static int process_command(AVFilterContext *ctx,
381 const char *cmd,
382 const char *arg,
383 char *res,
384 int res_len,
385 int flags)
386{
387 int ret = ff_filter_process_command(ctx, cmd, arg, res, res_len, flags);
388
389 if (ret < 0)
390 return ret;
391
392 return 0;
393}
394
395static const AVFilterPad guided_inputs[] = {
396 {
397 .name = "main",
398 .type = AVMEDIA_TYPE_VIDEO,
399 },{
400 .name = "reference",
401 .type = AVMEDIA_TYPE_VIDEO,
402 .config_props = config_input,
403 },
404 { NULL }
405};
406
407static const AVFilterPad guided_outputs[] = {
408 {
409 .name = "default",
410 .type = AVMEDIA_TYPE_VIDEO,
411 .config_props = config_output,
412 },
413 { NULL }
414};
415
416AVFilter ff_vf_guided = {
417 .name = "guided",
418 .description = NULL_IF_CONFIG_SMALL("Apply Guided filter."),
419 .init = init,
420 .uninit = uninit,
421 .query_formats = query_formats,
422 .priv_size = sizeof(GuidedContext),
423 .priv_class = &guided_class,
424 .activate = activate,
425 .inputs = guided_inputs,
426 .outputs = guided_outputs,
427 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
428 .process_command = process_command,
429};

Keyboard shortcuts

General

Show keyboard shortcuts
?
Focus the current search
/

Issues and pull requests

Open filters
f

Issues, pull requests and Inbox

Select next item
j↓
Select previous item
k↑
Open selected item
Enter
Dismiss filters or clear selection
Esc

Repository code

Go to file
t