2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
12 * This is an example demonstrating multi-resolution encoding in VP8.
13 * High-resolution input video is down-sampled to lower-resolutions. The
14 * encoder then encodes the video and outputs multiple bitstreams with
15 * different resolutions.
22 #define VPX_CODEC_DISABLE_COMPAT 1
23 #include "vpx/vpx_encoder.h"
24 #include "vpx/vp8cx.h"
25 #include "vpx_ports/mem_ops.h"
26 #include "./tools_common.h"
27 #define interface (vpx_codec_vp8_cx())
28 #define fourcc 0x30385056
35 * The input video frame is downsampled several times to generate a multi-level
36 * hierarchical structure. NUM_ENCODERS is defined as the number of encoding
37 * levels required. For example, if the size of input video is 1280x720,
38 * NUM_ENCODERS is 3, and down-sampling factor is 2, the encoder outputs 3
39 * bitstreams with resolution of 1280x720(level 0), 640x360(level 1), and
40 * 320x180(level 2) respectively.
42 #define NUM_ENCODERS 3
44 /* This example uses the scaler function in libyuv. */
45 #include "third_party/libyuv/include/libyuv/basic_types.h"
46 #include "third_party/libyuv/include/libyuv/scale.h"
47 #include "third_party/libyuv/include/libyuv/cpu_id.h"
49 int (*read_frame_p)(FILE *f, vpx_image_t *img);
51 static int read_frame(FILE *f, vpx_image_t *img) {
52 size_t nbytes, to_read;
55 to_read = img->w*img->h*3/2;
56 nbytes = fread(img->planes[0], 1, to_read, f);
57 if(nbytes != to_read) {
60 printf("Warning: Read partial frame. Check your width & height!\n");
65 static int read_frame_by_row(FILE *f, vpx_image_t *img) {
66 size_t nbytes, to_read;
70 for (plane = 0; plane < 3; plane++)
73 int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
74 int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
77 /* Determine the correct plane based on the image format. The for-loop
78 * always counts in Y,U,V order, but this may not match the order of
84 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U];
87 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V];
90 ptr = img->planes[plane];
93 for (r = 0; r < h; r++)
97 nbytes = fread(ptr, 1, to_read, f);
98 if(nbytes != to_read) {
101 printf("Warning: Read partial frame. Check your width & height!\n");
105 ptr += img->stride[plane];
114 static void write_ivf_file_header(FILE *outfile,
115 const vpx_codec_enc_cfg_t *cfg,
119 if(cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS)
125 mem_put_le16(header+4, 0); /* version */
126 mem_put_le16(header+6, 32); /* headersize */
127 mem_put_le32(header+8, fourcc); /* headersize */
128 mem_put_le16(header+12, cfg->g_w); /* width */
129 mem_put_le16(header+14, cfg->g_h); /* height */
130 mem_put_le32(header+16, cfg->g_timebase.den); /* rate */
131 mem_put_le32(header+20, cfg->g_timebase.num); /* scale */
132 mem_put_le32(header+24, frame_cnt); /* length */
133 mem_put_le32(header+28, 0); /* unused */
135 (void) fwrite(header, 1, 32, outfile);
138 static void write_ivf_frame_header(FILE *outfile,
139 const vpx_codec_cx_pkt_t *pkt)
144 if(pkt->kind != VPX_CODEC_CX_FRAME_PKT)
147 pts = pkt->data.frame.pts;
148 mem_put_le32(header, pkt->data.frame.sz);
149 mem_put_le32(header+4, pts&0xFFFFFFFF);
150 mem_put_le32(header+8, pts >> 32);
152 (void) fwrite(header, 1, 12, outfile);
155 int main(int argc, char **argv)
157 FILE *infile, *outfile[NUM_ENCODERS];
158 vpx_codec_ctx_t codec[NUM_ENCODERS];
159 vpx_codec_enc_cfg_t cfg[NUM_ENCODERS];
160 vpx_codec_pts_t frame_cnt = 0;
161 vpx_image_t raw[NUM_ENCODERS];
162 vpx_codec_err_t res[NUM_ENCODERS];
171 /*Currently, only realtime mode is supported in multi-resolution encoding.*/
172 int arg_deadline = VPX_DL_REALTIME;
174 /* Set show_psnr to 1/0 to show/not show PSNR. Choose show_psnr=0 if you
175 don't need to know PSNR, which will skip PSNR calculation and save
178 uint64_t psnr_sse_total[NUM_ENCODERS] = {0};
179 uint64_t psnr_samples_total[NUM_ENCODERS] = {0};
180 double psnr_totals[NUM_ENCODERS][4] = {{0,0}};
181 int psnr_count[NUM_ENCODERS] = {0};
183 /* Set the required target bitrates for each resolution level.
184 * If target bitrate for highest-resolution level is set to 0,
185 * (i.e. target_bitrate[0]=0), we skip encoding at that level.
187 unsigned int target_bitrate[NUM_ENCODERS]={1000, 500, 100};
188 /* Enter the frame rate of the input video */
190 /* Set down-sampling factor for each resolution level.
191 dsf[0] controls down sampling from level 0 to level 1;
192 dsf[1] controls down sampling from level 1 to level 2;
193 dsf[2] is not used. */
194 vpx_rational_t dsf[NUM_ENCODERS] = {{2, 1}, {2, 1}, {1, 1}};
196 if(argc!= (5+NUM_ENCODERS))
197 die("Usage: %s <width> <height> <infile> <outfile(s)> <output psnr?>\n",
200 printf("Using %s\n",vpx_codec_iface_name(interface));
202 width = strtol(argv[1], NULL, 0);
203 height = strtol(argv[2], NULL, 0);
205 if(width < 16 || width%2 || height <16 || height%2)
206 die("Invalid resolution: %ldx%ld", width, height);
208 /* Open input video file for encoding */
209 if(!(infile = fopen(argv[3], "rb")))
210 die("Failed to open %s for reading", argv[3]);
212 /* Open output file for each encoder to output bitstreams */
213 for (i=0; i< NUM_ENCODERS; i++)
215 if(!target_bitrate[i])
221 if(!(outfile[i] = fopen(argv[i+4], "wb")))
222 die("Failed to open %s for writing", argv[i+4]);
225 show_psnr = strtol(argv[NUM_ENCODERS + 4], NULL, 0);
227 /* Populate default encoder configuration */
228 for (i=0; i< NUM_ENCODERS; i++)
230 res[i] = vpx_codec_enc_config_default(interface, &cfg[i], 0);
232 printf("Failed to get config: %s\n", vpx_codec_err_to_string(res[i]));
238 * Update the default configuration according to needs of the application.
240 /* Highest-resolution encoder settings */
243 cfg[0].g_threads = 1; /* number of threads used */
244 cfg[0].rc_dropframe_thresh = 30;
245 cfg[0].rc_end_usage = VPX_CBR;
246 cfg[0].rc_resize_allowed = 0;
247 cfg[0].rc_min_quantizer = 4;
248 cfg[0].rc_max_quantizer = 56;
249 cfg[0].rc_undershoot_pct = 98;
250 cfg[0].rc_overshoot_pct = 100;
251 cfg[0].rc_buf_initial_sz = 500;
252 cfg[0].rc_buf_optimal_sz = 600;
253 cfg[0].rc_buf_sz = 1000;
254 cfg[0].g_error_resilient = 1; /* Enable error resilient mode */
255 cfg[0].g_lag_in_frames = 0;
257 /* Disable automatic keyframe placement */
258 /* Note: These 3 settings are copied to all levels. But, except the lowest
259 * resolution level, all other levels are set to VPX_KF_DISABLED internally.
261 //cfg[0].kf_mode = VPX_KF_DISABLED;
262 cfg[0].kf_mode = VPX_KF_AUTO;
263 cfg[0].kf_min_dist = 3000;
264 cfg[0].kf_max_dist = 3000;
266 cfg[0].rc_target_bitrate = target_bitrate[0]; /* Set target bitrate */
267 cfg[0].g_timebase.num = 1; /* Set fps */
268 cfg[0].g_timebase.den = framerate;
270 /* Other-resolution encoder settings */
271 for (i=1; i< NUM_ENCODERS; i++)
273 memcpy(&cfg[i], &cfg[0], sizeof(vpx_codec_enc_cfg_t));
275 cfg[i].g_threads = 1; /* number of threads used */
276 cfg[i].rc_target_bitrate = target_bitrate[i];
278 /* Note: Width & height of other-resolution encoders are calculated
279 * from the highest-resolution encoder's size and the corresponding
280 * down_sampling_factor.
283 unsigned int iw = cfg[i-1].g_w*dsf[i-1].den + dsf[i-1].num - 1;
284 unsigned int ih = cfg[i-1].g_h*dsf[i-1].den + dsf[i-1].num - 1;
285 cfg[i].g_w = iw/dsf[i-1].num;
286 cfg[i].g_h = ih/dsf[i-1].num;
289 /* Make width & height to be multiplier of 2. */
290 // Should support odd size ???
291 if((cfg[i].g_w)%2)cfg[i].g_w++;
292 if((cfg[i].g_h)%2)cfg[i].g_h++;
295 /* Allocate image for each encoder */
296 for (i=0; i< NUM_ENCODERS; i++)
297 if(!vpx_img_alloc(&raw[i], VPX_IMG_FMT_I420, cfg[i].g_w, cfg[i].g_h, 32))
298 die("Failed to allocate image", cfg[i].g_w, cfg[i].g_h);
300 if (raw[0].stride[VPX_PLANE_Y] == raw[0].d_w)
301 read_frame_p = read_frame;
303 read_frame_p = read_frame_by_row;
305 for (i=0; i< NUM_ENCODERS; i++)
307 write_ivf_file_header(outfile[i], &cfg[i], 0);
309 /* Initialize multi-encoder */
310 if(vpx_codec_enc_init_multi(&codec[0], interface, &cfg[0], NUM_ENCODERS,
311 (show_psnr ? VPX_CODEC_USE_PSNR : 0), &dsf[0]))
312 die_codec(&codec[0], "Failed to initialize encoder");
314 /* The extra encoding configuration parameters can be set as follows. */
315 /* Set encoding speed */
316 for ( i=0; i<NUM_ENCODERS; i++)
319 if(vpx_codec_control(&codec[i], VP8E_SET_CPUUSED, speed))
320 die_codec(&codec[i], "Failed to set cpu_used");
323 /* Set static threshold. */
324 for ( i=0; i<NUM_ENCODERS; i++)
326 unsigned int static_thresh = 1;
327 if(vpx_codec_control(&codec[i], VP8E_SET_STATIC_THRESHOLD, static_thresh))
328 die_codec(&codec[i], "Failed to set static threshold");
331 /* Set NOISE_SENSITIVITY to do TEMPORAL_DENOISING */
332 /* Enable denoising for the highest-resolution encoder. */
333 if(vpx_codec_control(&codec[0], VP8E_SET_NOISE_SENSITIVITY, 1))
334 die_codec(&codec[0], "Failed to set noise_sensitivity");
335 for ( i=1; i< NUM_ENCODERS; i++)
337 if(vpx_codec_control(&codec[i], VP8E_SET_NOISE_SENSITIVITY, 0))
338 die_codec(&codec[i], "Failed to set noise_sensitivity");
345 while(frame_avail || got_data)
347 vpx_codec_iter_t iter[NUM_ENCODERS]={NULL};
348 const vpx_codec_cx_pkt_t *pkt[NUM_ENCODERS];
351 frame_avail = read_frame_p(infile, &raw[0]);
355 for ( i=1; i<NUM_ENCODERS; i++)
357 /*Scale the image down a number of times by downsampling factor*/
358 /* FilterMode 1 or 2 give better psnr than FilterMode 0. */
359 I420Scale(raw[i-1].planes[VPX_PLANE_Y], raw[i-1].stride[VPX_PLANE_Y],
360 raw[i-1].planes[VPX_PLANE_U], raw[i-1].stride[VPX_PLANE_U],
361 raw[i-1].planes[VPX_PLANE_V], raw[i-1].stride[VPX_PLANE_V],
362 raw[i-1].d_w, raw[i-1].d_h,
363 raw[i].planes[VPX_PLANE_Y], raw[i].stride[VPX_PLANE_Y],
364 raw[i].planes[VPX_PLANE_U], raw[i].stride[VPX_PLANE_U],
365 raw[i].planes[VPX_PLANE_V], raw[i].stride[VPX_PLANE_V],
366 raw[i].d_w, raw[i].d_h, 1);
370 /* Encode each frame at multi-levels */
371 if(vpx_codec_encode(&codec[0], frame_avail? &raw[0] : NULL,
372 frame_cnt, 1, flags, arg_deadline))
373 die_codec(&codec[0], "Failed to encode frame");
375 for (i=NUM_ENCODERS-1; i>=0 ; i--)
379 while( (pkt[i] = vpx_codec_get_cx_data(&codec[i], &iter[i])) )
382 switch(pkt[i]->kind) {
383 case VPX_CODEC_CX_FRAME_PKT:
384 write_ivf_frame_header(outfile[i], pkt[i]);
385 (void) fwrite(pkt[i]->data.frame.buf, 1,
386 pkt[i]->data.frame.sz, outfile[i]);
388 case VPX_CODEC_PSNR_PKT:
393 psnr_sse_total[i] += pkt[i]->data.psnr.sse[0];
394 psnr_samples_total[i] += pkt[i]->data.psnr.samples[0];
395 for (j = 0; j < 4; j++)
397 //fprintf(stderr, "%.3lf ", pkt[i]->data.psnr.psnr[j]);
398 psnr_totals[i][j] += pkt[i]->data.psnr.psnr[j];
407 printf(pkt[i]->kind == VPX_CODEC_CX_FRAME_PKT
408 && (pkt[i]->data.frame.flags & VPX_FRAME_IS_KEY)? "K":".");
418 printf("Processed %ld frames.\n",(long int)frame_cnt-1);
419 for (i=0; i< NUM_ENCODERS; i++)
421 /* Calculate PSNR and print it out */
422 if ( (show_psnr) && (psnr_count[i]>0) )
425 double ovpsnr = sse_to_psnr(psnr_samples_total[i], 255.0,
428 fprintf(stderr, "\n ENC%d PSNR (Overall/Avg/Y/U/V)", i);
430 fprintf(stderr, " %.3lf", ovpsnr);
431 for (j = 0; j < 4; j++)
433 fprintf(stderr, " %.3lf", psnr_totals[i][j]/psnr_count[i]);
437 if(vpx_codec_destroy(&codec[i]))
438 die_codec(&codec[i], "Failed to destroy codec");
440 vpx_img_free(&raw[i]);
445 /* Try to rewrite the file header with the actual frame count */
446 if(!fseek(outfile[i], 0, SEEK_SET))
447 write_ivf_file_header(outfile[i], &cfg[i], frame_cnt-1);