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 #define interface (vpx_codec_vp8_cx()) 27 #define fourcc 0x30385056 28 29 #define IVF_FILE_HDR_SZ (32) 30 #define IVF_FRAME_HDR_SZ (12) 31 32 /* 33 * The input video frame is downsampled several times to generate a multi-level 34 * hierarchical structure. NUM_ENCODERS is defined as the number of encoding 35 * levels required. For example, if the size of input video is 1280x720, 36 * NUM_ENCODERS is 3, and down-sampling factor is 2, the encoder outputs 3 37 * bitstreams with resolution of 1280x720(level 0), 640x360(level 1), and 38 * 320x180(level 2) respectively. 39 */ 40 #define NUM_ENCODERS 3 41 42 /* This example uses the scaler function in libyuv. */ 43 #include "third_party/libyuv/include/libyuv/basic_types.h" 44 #include "third_party/libyuv/include/libyuv/scale.h" 45 #include "third_party/libyuv/include/libyuv/cpu_id.h" 46 47 static double vp8_mse2psnr(double Samples, double Peak, double Mse) 48 { 49 double psnr; 50 51 if ((double)Mse > 0.0) 52 psnr = 10.0 * log10(Peak * Peak * Samples / Mse); 53 else 54 psnr = 60; // Limit to prevent / 0 55 56 if (psnr > 60) 57 psnr = 60; 58 59 return psnr; 60 } 61 62 static void die(const char *fmt, ...) { 63 va_list ap; 64 65 va_start(ap, fmt); 66 vprintf(fmt, ap); 67 if(fmt[strlen(fmt)-1] != '\n') 68 printf("\n"); 69 exit(EXIT_FAILURE); 70 } 71 72 static void die_codec(vpx_codec_ctx_t *ctx, const char *s) { 73 const char *detail = vpx_codec_error_detail(ctx); 74 75 printf("%s: %s\n", s, vpx_codec_error(ctx)); 76 if(detail) 77 printf(" %s\n",detail); 78 exit(EXIT_FAILURE); 79 } 80 81 int (*read_frame_p)(FILE *f, vpx_image_t *img); 82 83 static int read_frame(FILE *f, vpx_image_t *img) { 84 size_t nbytes, to_read; 85 int res = 1; 86 87 to_read = img->w*img->h*3/2; 88 nbytes = fread(img->planes[0], 1, to_read, f); 89 if(nbytes != to_read) { 90 res = 0; 91 if(nbytes > 0) 92 printf("Warning: Read partial frame. Check your width & height!\n"); 93 } 94 return res; 95 } 96 97 static int read_frame_by_row(FILE *f, vpx_image_t *img) { 98 size_t nbytes, to_read; 99 int res = 1; 100 int plane; 101 102 for (plane = 0; plane < 3; plane++) 103 { 104 unsigned char *ptr; 105 int w = (plane ? (1 + img->d_w) / 2 : img->d_w); 106 int h = (plane ? (1 + img->d_h) / 2 : img->d_h); 107 int r; 108 109 /* Determine the correct plane based on the image format. The for-loop 110 * always counts in Y,U,V order, but this may not match the order of 111 * the data on disk. 112 */ 113 switch (plane) 114 { 115 case 1: 116 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U]; 117 break; 118 case 2: 119 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V]; 120 break; 121 default: 122 ptr = img->planes[plane]; 123 } 124 125 for (r = 0; r < h; r++) 126 { 127 to_read = w; 128 129 nbytes = fread(ptr, 1, to_read, f); 130 if(nbytes != to_read) { 131 res = 0; 132 if(nbytes > 0) 133 printf("Warning: Read partial frame. Check your width & height!\n"); 134 break; 135 } 136 137 ptr += img->stride[plane]; 138 } 139 if (!res) 140 break; 141 } 142 143 return res; 144 } 145 146 static void write_ivf_file_header(FILE *outfile, 147 const vpx_codec_enc_cfg_t *cfg, 148 int frame_cnt) { 149 char header[32]; 150 151 if(cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS) 152 return; 153 header[0] = 'D'; 154 header[1] = 'K'; 155 header[2] = 'I'; 156 header[3] = 'F'; 157 mem_put_le16(header+4, 0); /* version */ 158 mem_put_le16(header+6, 32); /* headersize */ 159 mem_put_le32(header+8, fourcc); /* headersize */ 160 mem_put_le16(header+12, cfg->g_w); /* width */ 161 mem_put_le16(header+14, cfg->g_h); /* height */ 162 mem_put_le32(header+16, cfg->g_timebase.den); /* rate */ 163 mem_put_le32(header+20, cfg->g_timebase.num); /* scale */ 164 mem_put_le32(header+24, frame_cnt); /* length */ 165 mem_put_le32(header+28, 0); /* unused */ 166 167 (void) fwrite(header, 1, 32, outfile); 168 } 169 170 static void write_ivf_frame_header(FILE *outfile, 171 const vpx_codec_cx_pkt_t *pkt) 172 { 173 char header[12]; 174 vpx_codec_pts_t pts; 175 176 if(pkt->kind != VPX_CODEC_CX_FRAME_PKT) 177 return; 178 179 pts = pkt->data.frame.pts; 180 mem_put_le32(header, pkt->data.frame.sz); 181 mem_put_le32(header+4, pts&0xFFFFFFFF); 182 mem_put_le32(header+8, pts >> 32); 183 184 (void) fwrite(header, 1, 12, outfile); 185 } 186 187 int main(int argc, char **argv) 188 { 189 FILE *infile, *outfile[NUM_ENCODERS]; 190 vpx_codec_ctx_t codec[NUM_ENCODERS]; 191 vpx_codec_enc_cfg_t cfg[NUM_ENCODERS]; 192 vpx_codec_pts_t frame_cnt = 0; 193 vpx_image_t raw[NUM_ENCODERS]; 194 vpx_codec_err_t res[NUM_ENCODERS]; 195 196 int i; 197 long width; 198 long height; 199 int frame_avail; 200 int got_data; 201 int flags = 0; 202 203 /*Currently, only realtime mode is supported in multi-resolution encoding.*/ 204 int arg_deadline = VPX_DL_REALTIME; 205 206 /* Set show_psnr to 1/0 to show/not show PSNR. Choose show_psnr=0 if you 207 don't need to know PSNR, which will skip PSNR calculation and save 208 encoding time. */ 209 int show_psnr = 0; 210 uint64_t psnr_sse_total[NUM_ENCODERS] = {0}; 211 uint64_t psnr_samples_total[NUM_ENCODERS] = {0}; 212 double psnr_totals[NUM_ENCODERS][4] = {{0,0}}; 213 int psnr_count[NUM_ENCODERS] = {0}; 214 215 /* Set the required target bitrates for each resolution level. 216 * If target bitrate for highest-resolution level is set to 0, 217 * (i.e. target_bitrate[0]=0), we skip encoding at that level. 218 */ 219 unsigned int target_bitrate[NUM_ENCODERS]={1400, 500, 100}; 220 /* Enter the frame rate of the input video */ 221 int framerate = 30; 222 /* Set down-sampling factor for each resolution level. 223 dsf[0] controls down sampling from level 0 to level 1; 224 dsf[1] controls down sampling from level 1 to level 2; 225 dsf[2] is not used. */ 226 vpx_rational_t dsf[NUM_ENCODERS] = {{2, 1}, {2, 1}, {1, 1}}; 227 228 if(argc!= (5+NUM_ENCODERS)) 229 die("Usage: %s <width> <height> <infile> <outfile(s)> <output psnr?>\n", 230 argv[0]); 231 232 printf("Using %s\n",vpx_codec_iface_name(interface)); 233 234 width = strtol(argv[1], NULL, 0); 235 height = strtol(argv[2], NULL, 0); 236 237 if(width < 16 || width%2 || height <16 || height%2) 238 die("Invalid resolution: %ldx%ld", width, height); 239 240 /* Open input video file for encoding */ 241 if(!(infile = fopen(argv[3], "rb"))) 242 die("Failed to open %s for reading", argv[3]); 243 244 /* Open output file for each encoder to output bitstreams */ 245 for (i=0; i< NUM_ENCODERS; i++) 246 { 247 if(!target_bitrate[i]) 248 { 249 outfile[i] = NULL; 250 continue; 251 } 252 253 if(!(outfile[i] = fopen(argv[i+4], "wb"))) 254 die("Failed to open %s for writing", argv[i+4]); 255 } 256 257 show_psnr = strtol(argv[NUM_ENCODERS + 4], NULL, 0); 258 259 /* Populate default encoder configuration */ 260 for (i=0; i< NUM_ENCODERS; i++) 261 { 262 res[i] = vpx_codec_enc_config_default(interface, &cfg[i], 0); 263 if(res[i]) { 264 printf("Failed to get config: %s\n", vpx_codec_err_to_string(res[i])); 265 return EXIT_FAILURE; 266 } 267 } 268 269 /* 270 * Update the default configuration according to needs of the application. 271 */ 272 /* Highest-resolution encoder settings */ 273 cfg[0].g_w = width; 274 cfg[0].g_h = height; 275 cfg[0].g_threads = 1; /* number of threads used */ 276 cfg[0].rc_dropframe_thresh = 30; 277 cfg[0].rc_end_usage = VPX_CBR; 278 cfg[0].rc_resize_allowed = 0; 279 cfg[0].rc_min_quantizer = 4; 280 cfg[0].rc_max_quantizer = 56; 281 cfg[0].rc_undershoot_pct = 98; 282 cfg[0].rc_overshoot_pct = 100; 283 cfg[0].rc_buf_initial_sz = 500; 284 cfg[0].rc_buf_optimal_sz = 600; 285 cfg[0].rc_buf_sz = 1000; 286 cfg[0].g_error_resilient = 1; /* Enable error resilient mode */ 287 cfg[0].g_lag_in_frames = 0; 288 289 /* Disable automatic keyframe placement */ 290 /* Note: These 3 settings are copied to all levels. But, except the lowest 291 * resolution level, all other levels are set to VPX_KF_DISABLED internally. 292 */ 293 //cfg[0].kf_mode = VPX_KF_DISABLED; 294 cfg[0].kf_mode = VPX_KF_AUTO; 295 cfg[0].kf_min_dist = 3000; 296 cfg[0].kf_max_dist = 3000; 297 298 cfg[0].rc_target_bitrate = target_bitrate[0]; /* Set target bitrate */ 299 cfg[0].g_timebase.num = 1; /* Set fps */ 300 cfg[0].g_timebase.den = framerate; 301 302 /* Other-resolution encoder settings */ 303 for (i=1; i< NUM_ENCODERS; i++) 304 { 305 memcpy(&cfg[i], &cfg[0], sizeof(vpx_codec_enc_cfg_t)); 306 307 cfg[i].g_threads = 1; /* number of threads used */ 308 cfg[i].rc_target_bitrate = target_bitrate[i]; 309 310 /* Note: Width & height of other-resolution encoders are calculated 311 * from the highest-resolution encoder's size and the corresponding 312 * down_sampling_factor. 313 */ 314 { 315 unsigned int iw = cfg[i-1].g_w*dsf[i-1].den + dsf[i-1].num - 1; 316 unsigned int ih = cfg[i-1].g_h*dsf[i-1].den + dsf[i-1].num - 1; 317 cfg[i].g_w = iw/dsf[i-1].num; 318 cfg[i].g_h = ih/dsf[i-1].num; 319 } 320 321 /* Make width & height to be multiplier of 2. */ 322 // Should support odd size ??? 323 if((cfg[i].g_w)%2)cfg[i].g_w++; 324 if((cfg[i].g_h)%2)cfg[i].g_h++; 325 } 326 327 /* Allocate image for each encoder */ 328 for (i=0; i< NUM_ENCODERS; i++) 329 if(!vpx_img_alloc(&raw[i], VPX_IMG_FMT_I420, cfg[i].g_w, cfg[i].g_h, 32)) 330 die("Failed to allocate image", cfg[i].g_w, cfg[i].g_h); 331 332 if (raw[0].stride[VPX_PLANE_Y] == raw[0].d_w) 333 read_frame_p = read_frame; 334 else 335 read_frame_p = read_frame_by_row; 336 337 for (i=0; i< NUM_ENCODERS; i++) 338 if(outfile[i]) 339 write_ivf_file_header(outfile[i], &cfg[i], 0); 340 341 /* Initialize multi-encoder */ 342 if(vpx_codec_enc_init_multi(&codec[0], interface, &cfg[0], NUM_ENCODERS, 343 (show_psnr ? VPX_CODEC_USE_PSNR : 0), &dsf[0])) 344 die_codec(&codec[0], "Failed to initialize encoder"); 345 346 /* The extra encoding configuration parameters can be set as follows. */ 347 /* Set encoding speed */ 348 for ( i=0; i<NUM_ENCODERS; i++) 349 { 350 int speed = -6; 351 if(vpx_codec_control(&codec[i], VP8E_SET_CPUUSED, speed)) 352 die_codec(&codec[i], "Failed to set cpu_used"); 353 } 354 /* Set static thresh for highest-resolution encoder. Set it to 1000 for 355 * better performance. */ 356 { 357 unsigned int static_thresh = 1000; 358 if(vpx_codec_control(&codec[0], VP8E_SET_STATIC_THRESHOLD, static_thresh)) 359 die_codec(&codec[0], "Failed to set static threshold"); 360 } 361 /* Set static thresh = 0 for other encoders for better quality */ 362 for ( i=1; i<NUM_ENCODERS; i++) 363 { 364 unsigned int static_thresh = 0; 365 if(vpx_codec_control(&codec[i], VP8E_SET_STATIC_THRESHOLD, static_thresh)) 366 die_codec(&codec[i], "Failed to set static threshold"); 367 } 368 /* Set NOISE_SENSITIVITY to do TEMPORAL_DENOISING */ 369 for ( i=0; i< NUM_ENCODERS; i++) 370 { 371 if(vpx_codec_control(&codec[i], VP8E_SET_NOISE_SENSITIVITY, 0)) 372 die_codec(&codec[i], "Failed to set noise_sensitivity"); 373 } 374 375 frame_avail = 1; 376 got_data = 0; 377 378 while(frame_avail || got_data) 379 { 380 vpx_codec_iter_t iter[NUM_ENCODERS]={NULL}; 381 const vpx_codec_cx_pkt_t *pkt[NUM_ENCODERS]; 382 383 flags = 0; 384 frame_avail = read_frame_p(infile, &raw[0]); 385 386 if(frame_avail) 387 { 388 for ( i=1; i<NUM_ENCODERS; i++) 389 { 390 /*Scale the image down a number of times by downsampling factor*/ 391 /* FilterMode 1 or 2 give better psnr than FilterMode 0. */ 392 I420Scale(raw[i-1].planes[VPX_PLANE_Y], raw[i-1].stride[VPX_PLANE_Y], 393 raw[i-1].planes[VPX_PLANE_U], raw[i-1].stride[VPX_PLANE_U], 394 raw[i-1].planes[VPX_PLANE_V], raw[i-1].stride[VPX_PLANE_V], 395 raw[i-1].d_w, raw[i-1].d_h, 396 raw[i].planes[VPX_PLANE_Y], raw[i].stride[VPX_PLANE_Y], 397 raw[i].planes[VPX_PLANE_U], raw[i].stride[VPX_PLANE_U], 398 raw[i].planes[VPX_PLANE_V], raw[i].stride[VPX_PLANE_V], 399 raw[i].d_w, raw[i].d_h, 1); 400 } 401 } 402 403 /* Encode each frame at multi-levels */ 404 if(vpx_codec_encode(&codec[0], frame_avail? &raw[0] : NULL, 405 frame_cnt, 1, flags, arg_deadline)) 406 die_codec(&codec[0], "Failed to encode frame"); 407 408 for (i=NUM_ENCODERS-1; i>=0 ; i--) 409 { 410 got_data = 0; 411 412 while( (pkt[i] = vpx_codec_get_cx_data(&codec[i], &iter[i])) ) 413 { 414 got_data = 1; 415 switch(pkt[i]->kind) { 416 case VPX_CODEC_CX_FRAME_PKT: 417 write_ivf_frame_header(outfile[i], pkt[i]); 418 (void) fwrite(pkt[i]->data.frame.buf, 1, 419 pkt[i]->data.frame.sz, outfile[i]); 420 break; 421 case VPX_CODEC_PSNR_PKT: 422 if (show_psnr) 423 { 424 int j; 425 426 psnr_sse_total[i] += pkt[i]->data.psnr.sse[0]; 427 psnr_samples_total[i] += pkt[i]->data.psnr.samples[0]; 428 for (j = 0; j < 4; j++) 429 { 430 //fprintf(stderr, "%.3lf ", pkt[i]->data.psnr.psnr[j]); 431 psnr_totals[i][j] += pkt[i]->data.psnr.psnr[j]; 432 } 433 psnr_count[i]++; 434 } 435 436 break; 437 default: 438 break; 439 } 440 printf(pkt[i]->kind == VPX_CODEC_CX_FRAME_PKT 441 && (pkt[i]->data.frame.flags & VPX_FRAME_IS_KEY)? "K":"."); 442 fflush(stdout); 443 } 444 } 445 frame_cnt++; 446 } 447 printf("\n"); 448 449 fclose(infile); 450 451 printf("Processed %ld frames.\n",(long int)frame_cnt-1); 452 for (i=0; i< NUM_ENCODERS; i++) 453 { 454 /* Calculate PSNR and print it out */ 455 if ( (show_psnr) && (psnr_count[i]>0) ) 456 { 457 int j; 458 double ovpsnr = vp8_mse2psnr(psnr_samples_total[i], 255.0, 459 psnr_sse_total[i]); 460 461 fprintf(stderr, "\n ENC%d PSNR (Overall/Avg/Y/U/V)", i); 462 463 fprintf(stderr, " %.3lf", ovpsnr); 464 for (j = 0; j < 4; j++) 465 { 466 fprintf(stderr, " %.3lf", psnr_totals[i][j]/psnr_count[i]); 467 } 468 } 469 470 if(vpx_codec_destroy(&codec[i])) 471 die_codec(&codec[i], "Failed to destroy codec"); 472 473 vpx_img_free(&raw[i]); 474 475 if(!outfile[i]) 476 continue; 477 478 /* Try to rewrite the file header with the actual frame count */ 479 if(!fseek(outfile[i], 0, SEEK_SET)) 480 write_ivf_file_header(outfile[i], &cfg[i], frame_cnt-1); 481 fclose(outfile[i]); 482 } 483 printf("\n"); 484 485 return EXIT_SUCCESS; 486 } 487