2 * Copyright (c) 2017 Ronald S. Bultje <rsbultje@gmail.com>
3 * Copyright (c) 2017 Ashish Pratap Singh <ashk43712@gmail.com>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * Calculate the VMAF between two input videos.
29 #include "libavutil/avstring.h"
30 #include "libavutil/opt.h"
31 #include "libavutil/pixdesc.h"
33 #include "drawutils.h"
35 #include "framesync.h"
39 typedef struct LIBVMAFContext {
42 const AVPixFmtDescriptor *desc;
46 int vmaf_thread_created;
47 pthread_t vmaf_thread;
68 #define OFFSET(x) offsetof(LIBVMAFContext, x)
69 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
71 static const AVOption libvmaf_options[] = {
72 {"model_path", "Set the model to be used for computing vmaf.", OFFSET(model_path), AV_OPT_TYPE_STRING, {.str="/usr/local/share/model/vmaf_v0.6.1.pkl"}, 0, 1, FLAGS},
73 {"log_path", "Set the file path to be used to store logs.", OFFSET(log_path), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 1, FLAGS},
74 {"log_fmt", "Set the format of the log (xml or json).", OFFSET(log_fmt), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 1, FLAGS},
75 {"enable_transform", "Enables transform for computing vmaf.", OFFSET(enable_transform), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS},
76 {"phone_model", "Invokes the phone model that will generate higher VMAF scores.", OFFSET(phone_model), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS},
77 {"psnr", "Enables computing psnr along with vmaf.", OFFSET(psnr), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS},
78 {"ssim", "Enables computing ssim along with vmaf.", OFFSET(ssim), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS},
79 {"ms_ssim", "Enables computing ms-ssim along with vmaf.", OFFSET(ms_ssim), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS},
80 {"pool", "Set the pool method to be used for computing vmaf.", OFFSET(pool), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 1, FLAGS},
84 FRAMESYNC_DEFINE_CLASS(libvmaf, LIBVMAFContext, fs);
86 #define read_frame_fn(type, bits) \
87 static int read_frame_##bits##bit(float *ref_data, float *main_data, \
88 float *temp_data, int stride, void *ctx) \
90 LIBVMAFContext *s = (LIBVMAFContext *) ctx; \
93 pthread_mutex_lock(&s->lock); \
95 while (!s->frame_set && !s->eof) { \
96 pthread_cond_wait(&s->cond, &s->lock); \
100 int ref_stride = s->gref->linesize[0]; \
101 int main_stride = s->gmain->linesize[0]; \
103 const type *ref_ptr = (const type *) s->gref->data[0]; \
104 const type *main_ptr = (const type *) s->gmain->data[0]; \
106 float *ptr = ref_data; \
113 for (i = 0; i < h; i++) { \
114 for ( j = 0; j < w; j++) { \
115 ptr[j] = (float)ref_ptr[j]; \
117 ref_ptr += ref_stride / sizeof(*ref_ptr); \
118 ptr += stride / sizeof(*ptr); \
123 for (i = 0; i < h; i++) { \
124 for (j = 0; j < w; j++) { \
125 ptr[j] = (float)main_ptr[j]; \
127 main_ptr += main_stride / sizeof(*main_ptr); \
128 ptr += stride / sizeof(*ptr); \
132 ret = !s->frame_set; \
134 av_frame_unref(s->gref); \
135 av_frame_unref(s->gmain); \
138 pthread_cond_signal(&s->cond); \
139 pthread_mutex_unlock(&s->lock); \
148 read_frame_fn(uint8_t, 8);
149 read_frame_fn(uint16_t, 10);
151 static void compute_vmaf_score(LIBVMAFContext *s)
153 int (*read_frame)(float *ref_data, float *main_data, float *temp_data,
154 int stride, void *ctx);
157 if (s->desc->comp[0].depth <= 8) {
158 read_frame = read_frame_8bit;
160 read_frame = read_frame_10bit;
163 format = (char *) s->desc->name;
165 s->error = compute_vmaf(&s->vmaf_score, format, s->width, s->height,
166 read_frame, s, s->model_path, s->log_path,
167 s->log_fmt, 0, 0, s->enable_transform,
168 s->phone_model, s->psnr, s->ssim,
169 s->ms_ssim, s->pool);
172 static void *call_vmaf(void *ctx)
174 LIBVMAFContext *s = (LIBVMAFContext *) ctx;
175 compute_vmaf_score(s);
177 av_log(ctx, AV_LOG_INFO, "VMAF score: %f\n",s->vmaf_score);
179 pthread_mutex_lock(&s->lock);
180 pthread_cond_signal(&s->cond);
181 pthread_mutex_unlock(&s->lock);
187 static int do_vmaf(FFFrameSync *fs)
189 AVFilterContext *ctx = fs->parent;
190 LIBVMAFContext *s = ctx->priv;
191 AVFrame *master, *ref;
194 ret = ff_framesync_dualinput_get(fs, &master, &ref);
198 return ff_filter_frame(ctx->outputs[0], master);
200 pthread_mutex_lock(&s->lock);
202 while (s->frame_set && !s->error) {
203 pthread_cond_wait(&s->cond, &s->lock);
207 av_log(ctx, AV_LOG_ERROR,
208 "libvmaf encountered an error, check log for details\n");
209 pthread_mutex_unlock(&s->lock);
210 return AVERROR(EINVAL);
213 av_frame_ref(s->gref, ref);
214 av_frame_ref(s->gmain, master);
218 pthread_cond_signal(&s->cond);
219 pthread_mutex_unlock(&s->lock);
221 return ff_filter_frame(ctx->outputs[0], master);
224 static av_cold int init(AVFilterContext *ctx)
226 LIBVMAFContext *s = ctx->priv;
228 s->gref = av_frame_alloc();
229 s->gmain = av_frame_alloc();
232 s->vmaf_thread_created = 0;
233 pthread_mutex_init(&s->lock, NULL);
234 pthread_cond_init (&s->cond, NULL);
236 s->fs.on_event = do_vmaf;
240 static int query_formats(AVFilterContext *ctx)
242 static const enum AVPixelFormat pix_fmts[] = {
243 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV420P,
244 AV_PIX_FMT_YUV444P10LE, AV_PIX_FMT_YUV422P10LE, AV_PIX_FMT_YUV420P10LE,
248 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
250 return AVERROR(ENOMEM);
251 return ff_set_common_formats(ctx, fmts_list);
255 static int config_input_ref(AVFilterLink *inlink)
257 AVFilterContext *ctx = inlink->dst;
258 LIBVMAFContext *s = ctx->priv;
261 if (ctx->inputs[0]->w != ctx->inputs[1]->w ||
262 ctx->inputs[0]->h != ctx->inputs[1]->h) {
263 av_log(ctx, AV_LOG_ERROR, "Width and height of input videos must be same.\n");
264 return AVERROR(EINVAL);
266 if (ctx->inputs[0]->format != ctx->inputs[1]->format) {
267 av_log(ctx, AV_LOG_ERROR, "Inputs must be of same pixel format.\n");
268 return AVERROR(EINVAL);
271 s->desc = av_pix_fmt_desc_get(inlink->format);
272 s->width = ctx->inputs[0]->w;
273 s->height = ctx->inputs[0]->h;
275 th = pthread_create(&s->vmaf_thread, NULL, call_vmaf, (void *) s);
277 av_log(ctx, AV_LOG_ERROR, "Thread creation failed.\n");
278 return AVERROR(EINVAL);
280 s->vmaf_thread_created = 1;
285 static int config_output(AVFilterLink *outlink)
287 AVFilterContext *ctx = outlink->src;
288 LIBVMAFContext *s = ctx->priv;
289 AVFilterLink *mainlink = ctx->inputs[0];
292 ret = ff_framesync_init_dualinput(&s->fs, ctx);
295 outlink->w = mainlink->w;
296 outlink->h = mainlink->h;
297 outlink->time_base = mainlink->time_base;
298 outlink->sample_aspect_ratio = mainlink->sample_aspect_ratio;
299 outlink->frame_rate = mainlink->frame_rate;
300 if ((ret = ff_framesync_configure(&s->fs)) < 0)
306 static int activate(AVFilterContext *ctx)
308 LIBVMAFContext *s = ctx->priv;
309 return ff_framesync_activate(&s->fs);
312 static av_cold void uninit(AVFilterContext *ctx)
314 LIBVMAFContext *s = ctx->priv;
316 ff_framesync_uninit(&s->fs);
318 pthread_mutex_lock(&s->lock);
320 pthread_cond_signal(&s->cond);
321 pthread_mutex_unlock(&s->lock);
323 if (s->vmaf_thread_created)
325 pthread_join(s->vmaf_thread, NULL);
326 s->vmaf_thread_created = 0;
329 av_frame_free(&s->gref);
330 av_frame_free(&s->gmain);
332 pthread_mutex_destroy(&s->lock);
333 pthread_cond_destroy(&s->cond);
336 static const AVFilterPad libvmaf_inputs[] = {
339 .type = AVMEDIA_TYPE_VIDEO,
342 .type = AVMEDIA_TYPE_VIDEO,
343 .config_props = config_input_ref,
348 static const AVFilterPad libvmaf_outputs[] = {
351 .type = AVMEDIA_TYPE_VIDEO,
352 .config_props = config_output,
357 AVFilter ff_vf_libvmaf = {
359 .description = NULL_IF_CONFIG_SMALL("Calculate the VMAF between two video streams."),
360 .preinit = libvmaf_framesync_preinit,
363 .query_formats = query_formats,
364 .activate = activate,
365 .priv_size = sizeof(LIBVMAFContext),
366 .priv_class = &libvmaf_class,
367 .inputs = libvmaf_inputs,
368 .outputs = libvmaf_outputs,