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 pthread_t vmaf_thread;
66 #define OFFSET(x) offsetof(LIBVMAFContext, x)
67 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
69 static const AVOption libvmaf_options[] = {
70 {"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},
71 {"log_path", "Set the file path to be used to store logs.", OFFSET(log_path), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 1, FLAGS},
72 {"log_fmt", "Set the format of the log (xml or json).", OFFSET(log_fmt), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 1, FLAGS},
73 {"enable_transform", "Enables transform for computing vmaf.", OFFSET(enable_transform), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS},
74 {"phone_model", "Invokes the phone model that will generate higher VMAF scores.", OFFSET(phone_model), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS},
75 {"psnr", "Enables computing psnr along with vmaf.", OFFSET(psnr), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS},
76 {"ssim", "Enables computing ssim along with vmaf.", OFFSET(ssim), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS},
77 {"ms_ssim", "Enables computing ms-ssim along with vmaf.", OFFSET(ms_ssim), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS},
78 {"pool", "Set the pool method to be used for computing vmaf.", OFFSET(pool), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 1, FLAGS},
82 FRAMESYNC_DEFINE_CLASS(libvmaf, LIBVMAFContext, fs);
84 #define read_frame_fn(type, bits) \
85 static int read_frame_##bits##bit(float *ref_data, float *main_data, \
86 float *temp_data, int stride, void *ctx) \
88 LIBVMAFContext *s = (LIBVMAFContext *) ctx; \
91 pthread_mutex_lock(&s->lock); \
93 while (!s->frame_set && !s->eof) { \
94 pthread_cond_wait(&s->cond, &s->lock); \
98 int ref_stride = s->gref->linesize[0]; \
99 int main_stride = s->gmain->linesize[0]; \
101 const type *ref_ptr = (const type *) s->gref->data[0]; \
102 const type *main_ptr = (const type *) s->gmain->data[0]; \
104 float *ptr = ref_data; \
111 for (i = 0; i < h; i++) { \
112 for ( j = 0; j < w; j++) { \
113 ptr[j] = (float)ref_ptr[j]; \
115 ref_ptr += ref_stride / sizeof(*ref_ptr); \
116 ptr += stride / sizeof(*ptr); \
121 for (i = 0; i < h; i++) { \
122 for (j = 0; j < w; j++) { \
123 ptr[j] = (float)main_ptr[j]; \
125 main_ptr += main_stride / sizeof(*main_ptr); \
126 ptr += stride / sizeof(*ptr); \
130 ret = !s->frame_set; \
134 pthread_cond_signal(&s->cond); \
135 pthread_mutex_unlock(&s->lock); \
144 read_frame_fn(uint8_t, 8);
145 read_frame_fn(uint16_t, 10);
147 static void compute_vmaf_score(LIBVMAFContext *s)
149 int (*read_frame)(float *ref_data, float *main_data, float *temp_data,
150 int stride, void *ctx);
153 if (s->desc->comp[0].depth <= 8) {
154 read_frame = read_frame_8bit;
156 read_frame = read_frame_10bit;
159 format = (char *) s->desc->name;
161 s->vmaf_score = compute_vmaf(format, s->width, s->height, read_frame, s,
162 s->model_path, s->log_path, s->log_fmt, 0, 0,
163 s->enable_transform, s->phone_model, s->psnr,
164 s->ssim, s->ms_ssim, s->pool);
167 static void *call_vmaf(void *ctx)
169 LIBVMAFContext *s = (LIBVMAFContext *) ctx;
170 compute_vmaf_score(s);
171 av_log(ctx, AV_LOG_INFO, "VMAF score: %f\n",s->vmaf_score);
175 static int do_vmaf(FFFrameSync *fs)
177 AVFilterContext *ctx = fs->parent;
178 LIBVMAFContext *s = ctx->priv;
179 AVFrame *master, *ref;
182 ret = ff_framesync_dualinput_get(fs, &master, &ref);
186 return ff_filter_frame(ctx->outputs[0], master);
188 pthread_mutex_lock(&s->lock);
190 while (s->frame_set != 0) {
191 pthread_cond_wait(&s->cond, &s->lock);
194 av_frame_ref(s->gref, ref);
195 av_frame_ref(s->gmain, master);
199 pthread_cond_signal(&s->cond);
200 pthread_mutex_unlock(&s->lock);
202 return ff_filter_frame(ctx->outputs[0], master);
205 static av_cold int init(AVFilterContext *ctx)
207 LIBVMAFContext *s = ctx->priv;
209 s->gref = av_frame_alloc();
210 s->gmain = av_frame_alloc();
212 pthread_mutex_init(&s->lock, NULL);
213 pthread_cond_init (&s->cond, NULL);
215 s->fs.on_event = do_vmaf;
219 static int query_formats(AVFilterContext *ctx)
221 static const enum AVPixelFormat pix_fmts[] = {
222 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV420P,
223 AV_PIX_FMT_YUV444P10LE, AV_PIX_FMT_YUV422P10LE, AV_PIX_FMT_YUV420P10LE,
227 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
229 return AVERROR(ENOMEM);
230 return ff_set_common_formats(ctx, fmts_list);
234 static int config_input_ref(AVFilterLink *inlink)
236 AVFilterContext *ctx = inlink->dst;
237 LIBVMAFContext *s = ctx->priv;
240 if (ctx->inputs[0]->w != ctx->inputs[1]->w ||
241 ctx->inputs[0]->h != ctx->inputs[1]->h) {
242 av_log(ctx, AV_LOG_ERROR, "Width and height of input videos must be same.\n");
243 return AVERROR(EINVAL);
245 if (ctx->inputs[0]->format != ctx->inputs[1]->format) {
246 av_log(ctx, AV_LOG_ERROR, "Inputs must be of same pixel format.\n");
247 return AVERROR(EINVAL);
250 s->desc = av_pix_fmt_desc_get(inlink->format);
251 s->width = ctx->inputs[0]->w;
252 s->height = ctx->inputs[0]->h;
254 th = pthread_create(&s->vmaf_thread, NULL, call_vmaf, (void *) s);
256 av_log(ctx, AV_LOG_ERROR, "Thread creation failed.\n");
257 return AVERROR(EINVAL);
263 static int config_output(AVFilterLink *outlink)
265 AVFilterContext *ctx = outlink->src;
266 LIBVMAFContext *s = ctx->priv;
267 AVFilterLink *mainlink = ctx->inputs[0];
270 ret = ff_framesync_init_dualinput(&s->fs, ctx);
273 outlink->w = mainlink->w;
274 outlink->h = mainlink->h;
275 outlink->time_base = mainlink->time_base;
276 outlink->sample_aspect_ratio = mainlink->sample_aspect_ratio;
277 outlink->frame_rate = mainlink->frame_rate;
278 if ((ret = ff_framesync_configure(&s->fs)) < 0)
284 static int activate(AVFilterContext *ctx)
286 LIBVMAFContext *s = ctx->priv;
287 return ff_framesync_activate(&s->fs);
290 static av_cold void uninit(AVFilterContext *ctx)
292 LIBVMAFContext *s = ctx->priv;
294 ff_framesync_uninit(&s->fs);
296 pthread_mutex_lock(&s->lock);
298 pthread_cond_signal(&s->cond);
299 pthread_mutex_unlock(&s->lock);
301 pthread_join(s->vmaf_thread, NULL);
303 av_frame_free(&s->gref);
304 av_frame_free(&s->gmain);
306 pthread_mutex_destroy(&s->lock);
307 pthread_cond_destroy(&s->cond);
310 static const AVFilterPad libvmaf_inputs[] = {
313 .type = AVMEDIA_TYPE_VIDEO,
316 .type = AVMEDIA_TYPE_VIDEO,
317 .config_props = config_input_ref,
322 static const AVFilterPad libvmaf_outputs[] = {
325 .type = AVMEDIA_TYPE_VIDEO,
326 .config_props = config_output,
331 AVFilter ff_vf_libvmaf = {
333 .description = NULL_IF_CONFIG_SMALL("Calculate the VMAF between two video streams."),
334 .preinit = libvmaf_framesync_preinit,
337 .query_formats = query_formats,
338 .activate = activate,
339 .priv_size = sizeof(LIBVMAFContext),
340 .priv_class = &libvmaf_class,
341 .inputs = libvmaf_inputs,
342 .outputs = libvmaf_outputs,