An error occurred while loading the file. Please try again.
-
Ronald S. Bultje authored
This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
c456b35f
vp8_multi_resolution_encoder.c 16.13 KiB
/* * Copyright (c) 2010 The WebM project authors. All Rights Reserved. * * Use of this source code is governed by a BSD-style license * that can be found in the LICENSE file in the root of the source * tree. An additional intellectual property rights grant can be found * in the file PATENTS. All contributing project authors may * be found in the AUTHORS file in the root of the source tree. *//* * This is an example demonstrating multi-resolution encoding in VP8. * High-resolution input video is down-sampled to lower-resolutions. The * encoder then encodes the video and outputs multiple bitstreams with * different resolutions. */#include <stdio.h>#include <stdlib.h>#include <stdarg.h>#include <string.h>#include "math.h"#define VPX_CODEC_DISABLE_COMPAT 1#include "vpx/vpx_encoder.h"#include "vpx/vp8cx.h"#include "vpx_ports/mem_ops.h"#define interface (vpx_codec_vp8_cx())#define fourcc 0x30385056#define IVF_FILE_HDR_SZ (32)#define IVF_FRAME_HDR_SZ (12)/* * The input video frame is downsampled several times to generate a multi-level * hierarchical structure. NUM_ENCODERS is defined as the number of encoding * levels required. For example, if the size of input video is 1280x720, * NUM_ENCODERS is 3, and down-sampling factor is 2, the encoder outputs 3 * bitstreams with resolution of 1280x720(level 0), 640x360(level 1), and * 320x180(level 2) respectively. */#define NUM_ENCODERS 3/* This example uses the scaler function in libyuv. */#include "third_party/libyuv/include/libyuv/basic_types.h"#include "third_party/libyuv/include/libyuv/scale.h"#include "third_party/libyuv/include/libyuv/cpu_id.h"static double vp8_mse2psnr(double Samples, double Peak, double Mse){ double psnr; if ((double)Mse > 0.0) psnr = 10.0 * log10(Peak * Peak * Samples / Mse); else psnr = 60; // Limit to prevent / 0 if (psnr > 60) psnr = 60; return psnr;}static void die(const char *fmt, ...) { va_list ap; va_start(ap, fmt); vprintf(fmt, ap); if(fmt[strlen(fmt)-1] != '\n') printf("\n"); exit(EXIT_FAILURE);}7172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140
static void die_codec(vpx_codec_ctx_t *ctx, const char *s) {
const char *detail = vpx_codec_error_detail(ctx);
printf("%s: %s\n", s, vpx_codec_error(ctx));
if(detail)
printf(" %s\n",detail);
exit(EXIT_FAILURE);
}
int (*read_frame_p)(FILE *f, vpx_image_t *img);
static int read_frame(FILE *f, vpx_image_t *img) {
size_t nbytes, to_read;
int res = 1;
to_read = img->w*img->h*3/2;
nbytes = fread(img->planes[0], 1, to_read, f);
if(nbytes != to_read) {
res = 0;
if(nbytes > 0)
printf("Warning: Read partial frame. Check your width & height!\n");
}
return res;
}
static int read_frame_by_row(FILE *f, vpx_image_t *img) {
size_t nbytes, to_read;
int res = 1;
int plane;
for (plane = 0; plane < 3; plane++)
{
unsigned char *ptr;
int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
int r;
/* Determine the correct plane based on the image format. The for-loop
* always counts in Y,U,V order, but this may not match the order of
* the data on disk.
*/
switch (plane)
{
case 1:
ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U];
break;
case 2:
ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V];
break;
default:
ptr = img->planes[plane];
}
for (r = 0; r < h; r++)
{
to_read = w;
nbytes = fread(ptr, 1, to_read, f);
if(nbytes != to_read) {
res = 0;
if(nbytes > 0)
printf("Warning: Read partial frame. Check your width & height!\n");
break;
}
ptr += img->stride[plane];
}
if (!res)
break;
141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210
}
return res;
}
static void write_ivf_file_header(FILE *outfile,
const vpx_codec_enc_cfg_t *cfg,
int frame_cnt) {
char header[32];
if(cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS)
return;
header[0] = 'D';
header[1] = 'K';
header[2] = 'I';
header[3] = 'F';
mem_put_le16(header+4, 0); /* version */
mem_put_le16(header+6, 32); /* headersize */
mem_put_le32(header+8, fourcc); /* headersize */
mem_put_le16(header+12, cfg->g_w); /* width */
mem_put_le16(header+14, cfg->g_h); /* height */
mem_put_le32(header+16, cfg->g_timebase.den); /* rate */
mem_put_le32(header+20, cfg->g_timebase.num); /* scale */
mem_put_le32(header+24, frame_cnt); /* length */
mem_put_le32(header+28, 0); /* unused */
(void) fwrite(header, 1, 32, outfile);
}
static void write_ivf_frame_header(FILE *outfile,
const vpx_codec_cx_pkt_t *pkt)
{
char header[12];
vpx_codec_pts_t pts;
if(pkt->kind != VPX_CODEC_CX_FRAME_PKT)
return;
pts = pkt->data.frame.pts;
mem_put_le32(header, pkt->data.frame.sz);
mem_put_le32(header+4, pts&0xFFFFFFFF);
mem_put_le32(header+8, pts >> 32);
(void) fwrite(header, 1, 12, outfile);
}
int main(int argc, char **argv)
{
FILE *infile, *outfile[NUM_ENCODERS];
vpx_codec_ctx_t codec[NUM_ENCODERS];
vpx_codec_enc_cfg_t cfg[NUM_ENCODERS];
vpx_codec_pts_t frame_cnt = 0;
vpx_image_t raw[NUM_ENCODERS];
vpx_codec_err_t res[NUM_ENCODERS];
int i;
long width;
long height;
int frame_avail;
int got_data;
int flags = 0;
/*Currently, only realtime mode is supported in multi-resolution encoding.*/
int arg_deadline = VPX_DL_REALTIME;
/* Set show_psnr to 1/0 to show/not show PSNR. Choose show_psnr=0 if you
don't need to know PSNR, which will skip PSNR calculation and save
encoding time. */
int show_psnr = 0;
uint64_t psnr_sse_total[NUM_ENCODERS] = {0};
211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280
uint64_t psnr_samples_total[NUM_ENCODERS] = {0};
double psnr_totals[NUM_ENCODERS][4] = {{0,0}};
int psnr_count[NUM_ENCODERS] = {0};
/* Set the required target bitrates for each resolution level.
* If target bitrate for highest-resolution level is set to 0,
* (i.e. target_bitrate[0]=0), we skip encoding at that level.
*/
unsigned int target_bitrate[NUM_ENCODERS]={1000, 500, 100};
/* Enter the frame rate of the input video */
int framerate = 30;
/* Set down-sampling factor for each resolution level.
dsf[0] controls down sampling from level 0 to level 1;
dsf[1] controls down sampling from level 1 to level 2;
dsf[2] is not used. */
vpx_rational_t dsf[NUM_ENCODERS] = {{2, 1}, {2, 1}, {1, 1}};
if(argc!= (5+NUM_ENCODERS))
die("Usage: %s <width> <height> <infile> <outfile(s)> <output psnr?>\n",
argv[0]);
printf("Using %s\n",vpx_codec_iface_name(interface));
width = strtol(argv[1], NULL, 0);
height = strtol(argv[2], NULL, 0);
if(width < 16 || width%2 || height <16 || height%2)
die("Invalid resolution: %ldx%ld", width, height);
/* Open input video file for encoding */
if(!(infile = fopen(argv[3], "rb")))
die("Failed to open %s for reading", argv[3]);
/* Open output file for each encoder to output bitstreams */
for (i=0; i< NUM_ENCODERS; i++)
{
if(!target_bitrate[i])
{
outfile[i] = NULL;
continue;
}
if(!(outfile[i] = fopen(argv[i+4], "wb")))
die("Failed to open %s for writing", argv[i+4]);
}
show_psnr = strtol(argv[NUM_ENCODERS + 4], NULL, 0);
/* Populate default encoder configuration */
for (i=0; i< NUM_ENCODERS; i++)
{
res[i] = vpx_codec_enc_config_default(interface, &cfg[i], 0);
if(res[i]) {
printf("Failed to get config: %s\n", vpx_codec_err_to_string(res[i]));
return EXIT_FAILURE;
}
}
/*
* Update the default configuration according to needs of the application.
*/
/* Highest-resolution encoder settings */
cfg[0].g_w = width;
cfg[0].g_h = height;
cfg[0].g_threads = 1; /* number of threads used */
cfg[0].rc_dropframe_thresh = 30;
cfg[0].rc_end_usage = VPX_CBR;
cfg[0].rc_resize_allowed = 0;
cfg[0].rc_min_quantizer = 4;
cfg[0].rc_max_quantizer = 56;
281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350
cfg[0].rc_undershoot_pct = 98;
cfg[0].rc_overshoot_pct = 100;
cfg[0].rc_buf_initial_sz = 500;
cfg[0].rc_buf_optimal_sz = 600;
cfg[0].rc_buf_sz = 1000;
cfg[0].g_error_resilient = 1; /* Enable error resilient mode */
cfg[0].g_lag_in_frames = 0;
/* Disable automatic keyframe placement */
/* Note: These 3 settings are copied to all levels. But, except the lowest
* resolution level, all other levels are set to VPX_KF_DISABLED internally.
*/
//cfg[0].kf_mode = VPX_KF_DISABLED;
cfg[0].kf_mode = VPX_KF_AUTO;
cfg[0].kf_min_dist = 3000;
cfg[0].kf_max_dist = 3000;
cfg[0].rc_target_bitrate = target_bitrate[0]; /* Set target bitrate */
cfg[0].g_timebase.num = 1; /* Set fps */
cfg[0].g_timebase.den = framerate;
/* Other-resolution encoder settings */
for (i=1; i< NUM_ENCODERS; i++)
{
memcpy(&cfg[i], &cfg[0], sizeof(vpx_codec_enc_cfg_t));
cfg[i].g_threads = 1; /* number of threads used */
cfg[i].rc_target_bitrate = target_bitrate[i];
/* Note: Width & height of other-resolution encoders are calculated
* from the highest-resolution encoder's size and the corresponding
* down_sampling_factor.
*/
{
unsigned int iw = cfg[i-1].g_w*dsf[i-1].den + dsf[i-1].num - 1;
unsigned int ih = cfg[i-1].g_h*dsf[i-1].den + dsf[i-1].num - 1;
cfg[i].g_w = iw/dsf[i-1].num;
cfg[i].g_h = ih/dsf[i-1].num;
}
/* Make width & height to be multiplier of 2. */
// Should support odd size ???
if((cfg[i].g_w)%2)cfg[i].g_w++;
if((cfg[i].g_h)%2)cfg[i].g_h++;
}
/* Allocate image for each encoder */
for (i=0; i< NUM_ENCODERS; i++)
if(!vpx_img_alloc(&raw[i], VPX_IMG_FMT_I420, cfg[i].g_w, cfg[i].g_h, 32))
die("Failed to allocate image", cfg[i].g_w, cfg[i].g_h);
if (raw[0].stride[VPX_PLANE_Y] == raw[0].d_w)
read_frame_p = read_frame;
else
read_frame_p = read_frame_by_row;
for (i=0; i< NUM_ENCODERS; i++)
if(outfile[i])
write_ivf_file_header(outfile[i], &cfg[i], 0);
/* Initialize multi-encoder */
if(vpx_codec_enc_init_multi(&codec[0], interface, &cfg[0], NUM_ENCODERS,
(show_psnr ? VPX_CODEC_USE_PSNR : 0), &dsf[0]))
die_codec(&codec[0], "Failed to initialize encoder");
/* The extra encoding configuration parameters can be set as follows. */
/* Set encoding speed */
for ( i=0; i<NUM_ENCODERS; i++)
{
int speed = -6;
351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420
if(vpx_codec_control(&codec[i], VP8E_SET_CPUUSED, speed))
die_codec(&codec[i], "Failed to set cpu_used");
}
/* Set static threshold. */
for ( i=0; i<NUM_ENCODERS; i++)
{
unsigned int static_thresh = 1;
if(vpx_codec_control(&codec[i], VP8E_SET_STATIC_THRESHOLD, static_thresh))
die_codec(&codec[i], "Failed to set static threshold");
}
/* Set NOISE_SENSITIVITY to do TEMPORAL_DENOISING */
/* Enable denoising for the highest-resolution encoder. */
if(vpx_codec_control(&codec[0], VP8E_SET_NOISE_SENSITIVITY, 1))
die_codec(&codec[0], "Failed to set noise_sensitivity");
for ( i=1; i< NUM_ENCODERS; i++)
{
if(vpx_codec_control(&codec[i], VP8E_SET_NOISE_SENSITIVITY, 0))
die_codec(&codec[i], "Failed to set noise_sensitivity");
}
frame_avail = 1;
got_data = 0;
while(frame_avail || got_data)
{
vpx_codec_iter_t iter[NUM_ENCODERS]={NULL};
const vpx_codec_cx_pkt_t *pkt[NUM_ENCODERS];
flags = 0;
frame_avail = read_frame_p(infile, &raw[0]);
if(frame_avail)
{
for ( i=1; i<NUM_ENCODERS; i++)
{
/*Scale the image down a number of times by downsampling factor*/
/* FilterMode 1 or 2 give better psnr than FilterMode 0. */
I420Scale(raw[i-1].planes[VPX_PLANE_Y], raw[i-1].stride[VPX_PLANE_Y],
raw[i-1].planes[VPX_PLANE_U], raw[i-1].stride[VPX_PLANE_U],
raw[i-1].planes[VPX_PLANE_V], raw[i-1].stride[VPX_PLANE_V],
raw[i-1].d_w, raw[i-1].d_h,
raw[i].planes[VPX_PLANE_Y], raw[i].stride[VPX_PLANE_Y],
raw[i].planes[VPX_PLANE_U], raw[i].stride[VPX_PLANE_U],
raw[i].planes[VPX_PLANE_V], raw[i].stride[VPX_PLANE_V],
raw[i].d_w, raw[i].d_h, 1);
}
}
/* Encode each frame at multi-levels */
if(vpx_codec_encode(&codec[0], frame_avail? &raw[0] : NULL,
frame_cnt, 1, flags, arg_deadline))
die_codec(&codec[0], "Failed to encode frame");
for (i=NUM_ENCODERS-1; i>=0 ; i--)
{
got_data = 0;
while( (pkt[i] = vpx_codec_get_cx_data(&codec[i], &iter[i])) )
{
got_data = 1;
switch(pkt[i]->kind) {
case VPX_CODEC_CX_FRAME_PKT:
write_ivf_frame_header(outfile[i], pkt[i]);
(void) fwrite(pkt[i]->data.frame.buf, 1,
pkt[i]->data.frame.sz, outfile[i]);
break;
case VPX_CODEC_PSNR_PKT:
421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486
if (show_psnr)
{
int j;
psnr_sse_total[i] += pkt[i]->data.psnr.sse[0];
psnr_samples_total[i] += pkt[i]->data.psnr.samples[0];
for (j = 0; j < 4; j++)
{
//fprintf(stderr, "%.3lf ", pkt[i]->data.psnr.psnr[j]);
psnr_totals[i][j] += pkt[i]->data.psnr.psnr[j];
}
psnr_count[i]++;
}
break;
default:
break;
}
printf(pkt[i]->kind == VPX_CODEC_CX_FRAME_PKT
&& (pkt[i]->data.frame.flags & VPX_FRAME_IS_KEY)? "K":".");
fflush(stdout);
}
}
frame_cnt++;
}
printf("\n");
fclose(infile);
printf("Processed %ld frames.\n",(long int)frame_cnt-1);
for (i=0; i< NUM_ENCODERS; i++)
{
/* Calculate PSNR and print it out */
if ( (show_psnr) && (psnr_count[i]>0) )
{
int j;
double ovpsnr = vp8_mse2psnr(psnr_samples_total[i], 255.0,
psnr_sse_total[i]);
fprintf(stderr, "\n ENC%d PSNR (Overall/Avg/Y/U/V)", i);
fprintf(stderr, " %.3lf", ovpsnr);
for (j = 0; j < 4; j++)
{
fprintf(stderr, " %.3lf", psnr_totals[i][j]/psnr_count[i]);
}
}
if(vpx_codec_destroy(&codec[i]))
die_codec(&codec[i], "Failed to destroy codec");
vpx_img_free(&raw[i]);
if(!outfile[i])
continue;
/* Try to rewrite the file header with the actual frame count */
if(!fseek(outfile[i], 0, SEEK_SET))
write_ivf_file_header(outfile[i], &cfg[i], frame_cnt-1);
fclose(outfile[i]);
}
printf("\n");
return EXIT_SUCCESS;
}