1/* 2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved. 3 * 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. 9 */ 10 11/* 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. 16 */ 17#include <stdio.h> 18#include <stdlib.h> 19#include <stdarg.h> 20#include <string.h> 21#include <math.h> 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 29 30void usage_exit() { 31 exit(EXIT_FAILURE); 32} 33 34/* 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. 41 */ 42#define NUM_ENCODERS 3 43 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" 48 49int (*read_frame_p)(FILE *f, vpx_image_t *img); 50 51static int read_frame(FILE *f, vpx_image_t *img) { 52 size_t nbytes, to_read; 53 int res = 1; 54 55 to_read = img->w*img->h*3/2; 56 nbytes = fread(img->planes[0], 1, to_read, f); 57 if(nbytes != to_read) { 58 res = 0; 59 if(nbytes > 0) 60 printf("Warning: Read partial frame. Check your width & height!\n"); 61 } 62 return res; 63} 64 65static int read_frame_by_row(FILE *f, vpx_image_t *img) { 66 size_t nbytes, to_read; 67 int res = 1; 68 int plane; 69 70 for (plane = 0; plane < 3; plane++) 71 { 72 unsigned char *ptr; 73 int w = (plane ? (1 + img->d_w) / 2 : img->d_w); 74 int h = (plane ? (1 + img->d_h) / 2 : img->d_h); 75 int r; 76 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 79 * the data on disk. 80 */ 81 switch (plane) 82 { 83 case 1: 84 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U]; 85 break; 86 case 2: 87 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V]; 88 break; 89 default: 90 ptr = img->planes[plane]; 91 } 92 93 for (r = 0; r < h; r++) 94 { 95 to_read = w; 96 97 nbytes = fread(ptr, 1, to_read, f); 98 if(nbytes != to_read) { 99 res = 0; 100 if(nbytes > 0) 101 printf("Warning: Read partial frame. Check your width & height!\n"); 102 break; 103 } 104 105 ptr += img->stride[plane]; 106 } 107 if (!res) 108 break; 109 } 110 111 return res; 112} 113 114static void write_ivf_file_header(FILE *outfile, 115 const vpx_codec_enc_cfg_t *cfg, 116 int frame_cnt) { 117 char header[32]; 118 119 if(cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS) 120 return; 121 header[0] = 'D'; 122 header[1] = 'K'; 123 header[2] = 'I'; 124 header[3] = 'F'; 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 */ 134 135 (void) fwrite(header, 1, 32, outfile); 136} 137 138static void write_ivf_frame_header(FILE *outfile, 139 const vpx_codec_cx_pkt_t *pkt) 140{ 141 char header[12]; 142 vpx_codec_pts_t pts; 143 144 if(pkt->kind != VPX_CODEC_CX_FRAME_PKT) 145 return; 146 147 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); 151 152 (void) fwrite(header, 1, 12, outfile); 153} 154 155int main(int argc, char **argv) 156{ 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]; 163 164 int i; 165 long width; 166 long height; 167 int frame_avail; 168 int got_data; 169 int flags = 0; 170 171 /*Currently, only realtime mode is supported in multi-resolution encoding.*/ 172 int arg_deadline = VPX_DL_REALTIME; 173 174 /* 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 176 encoding time. */ 177 int show_psnr = 0; 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}; 182 183 /* 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. 186 */ 187 unsigned int target_bitrate[NUM_ENCODERS]={1000, 500, 100}; 188 /* Enter the frame rate of the input video */ 189 int framerate = 30; 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}}; 195 196 if(argc!= (5+NUM_ENCODERS)) 197 die("Usage: %s <width> <height> <infile> <outfile(s)> <output psnr?>\n", 198 argv[0]); 199 200 printf("Using %s\n",vpx_codec_iface_name(interface)); 201 202 width = strtol(argv[1], NULL, 0); 203 height = strtol(argv[2], NULL, 0); 204 205 if(width < 16 || width%2 || height <16 || height%2) 206 die("Invalid resolution: %ldx%ld", width, height); 207 208 /* Open input video file for encoding */ 209 if(!(infile = fopen(argv[3], "rb"))) 210 die("Failed to open %s for reading", argv[3]); 211 212 /* Open output file for each encoder to output bitstreams */ 213 for (i=0; i< NUM_ENCODERS; i++) 214 { 215 if(!target_bitrate[i]) 216 { 217 outfile[i] = NULL; 218 continue; 219 } 220 221 if(!(outfile[i] = fopen(argv[i+4], "wb"))) 222 die("Failed to open %s for writing", argv[i+4]); 223 } 224 225 show_psnr = strtol(argv[NUM_ENCODERS + 4], NULL, 0); 226 227 /* Populate default encoder configuration */ 228 for (i=0; i< NUM_ENCODERS; i++) 229 { 230 res[i] = vpx_codec_enc_config_default(interface, &cfg[i], 0); 231 if(res[i]) { 232 printf("Failed to get config: %s\n", vpx_codec_err_to_string(res[i])); 233 return EXIT_FAILURE; 234 } 235 } 236 237 /* 238 * Update the default configuration according to needs of the application. 239 */ 240 /* Highest-resolution encoder settings */ 241 cfg[0].g_w = width; 242 cfg[0].g_h = height; 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; 256 257 /* 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. 260 */ 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; 265 266 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; 269 270 /* Other-resolution encoder settings */ 271 for (i=1; i< NUM_ENCODERS; i++) 272 { 273 memcpy(&cfg[i], &cfg[0], sizeof(vpx_codec_enc_cfg_t)); 274 275 cfg[i].g_threads = 1; /* number of threads used */ 276 cfg[i].rc_target_bitrate = target_bitrate[i]; 277 278 /* Note: Width & height of other-resolution encoders are calculated 279 * from the highest-resolution encoder's size and the corresponding 280 * down_sampling_factor. 281 */ 282 { 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; 287 } 288 289 /* 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++; 293 } 294 295 /* 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); 299 300 if (raw[0].stride[VPX_PLANE_Y] == raw[0].d_w) 301 read_frame_p = read_frame; 302 else 303 read_frame_p = read_frame_by_row; 304 305 for (i=0; i< NUM_ENCODERS; i++) 306 if(outfile[i]) 307 write_ivf_file_header(outfile[i], &cfg[i], 0); 308 309 /* 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"); 313 314 /* The extra encoding configuration parameters can be set as follows. */ 315 /* Set encoding speed */ 316 for ( i=0; i<NUM_ENCODERS; i++) 317 { 318 int speed = -6; 319 if(vpx_codec_control(&codec[i], VP8E_SET_CPUUSED, speed)) 320 die_codec(&codec[i], "Failed to set cpu_used"); 321 } 322 323 /* Set static threshold. */ 324 for ( i=0; i<NUM_ENCODERS; i++) 325 { 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"); 329 } 330 331 /* 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++) 336 { 337 if(vpx_codec_control(&codec[i], VP8E_SET_NOISE_SENSITIVITY, 0)) 338 die_codec(&codec[i], "Failed to set noise_sensitivity"); 339 } 340 341 342 frame_avail = 1; 343 got_data = 0; 344 345 while(frame_avail || got_data) 346 { 347 vpx_codec_iter_t iter[NUM_ENCODERS]={NULL}; 348 const vpx_codec_cx_pkt_t *pkt[NUM_ENCODERS]; 349 350 flags = 0; 351 frame_avail = read_frame_p(infile, &raw[0]); 352 353 if(frame_avail) 354 { 355 for ( i=1; i<NUM_ENCODERS; i++) 356 { 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); 367 } 368 } 369 370 /* 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"); 374 375 for (i=NUM_ENCODERS-1; i>=0 ; i--) 376 { 377 got_data = 0; 378 379 while( (pkt[i] = vpx_codec_get_cx_data(&codec[i], &iter[i])) ) 380 { 381 got_data = 1; 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]); 387 break; 388 case VPX_CODEC_PSNR_PKT: 389 if (show_psnr) 390 { 391 int j; 392 393 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++) 396 { 397 //fprintf(stderr, "%.3lf ", pkt[i]->data.psnr.psnr[j]); 398 psnr_totals[i][j] += pkt[i]->data.psnr.psnr[j]; 399 } 400 psnr_count[i]++; 401 } 402 403 break; 404 default: 405 break; 406 } 407 printf(pkt[i]->kind == VPX_CODEC_CX_FRAME_PKT 408 && (pkt[i]->data.frame.flags & VPX_FRAME_IS_KEY)? "K":"."); 409 fflush(stdout); 410 } 411 } 412 frame_cnt++; 413 } 414 printf("\n"); 415 416 fclose(infile); 417 418 printf("Processed %ld frames.\n",(long int)frame_cnt-1); 419 for (i=0; i< NUM_ENCODERS; i++) 420 { 421 /* Calculate PSNR and print it out */ 422 if ( (show_psnr) && (psnr_count[i]>0) ) 423 { 424 int j; 425 double ovpsnr = sse_to_psnr(psnr_samples_total[i], 255.0, 426 psnr_sse_total[i]); 427 428 fprintf(stderr, "\n ENC%d PSNR (Overall/Avg/Y/U/V)", i); 429 430 fprintf(stderr, " %.3lf", ovpsnr); 431 for (j = 0; j < 4; j++) 432 { 433 fprintf(stderr, " %.3lf", psnr_totals[i][j]/psnr_count[i]); 434 } 435 } 436 437 if(vpx_codec_destroy(&codec[i])) 438 die_codec(&codec[i], "Failed to destroy codec"); 439 440 vpx_img_free(&raw[i]); 441 442 if(!outfile[i]) 443 continue; 444 445 /* 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); 448 fclose(outfile[i]); 449 } 450 printf("\n"); 451 452 return EXIT_SUCCESS; 453} 454