WebM Codec SDK
vp9cx_set_ref
1/*
2 * Copyright (c) 2016 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// VP9 Set Reference Frame
12// ============================
13//
14// This is an example demonstrating how to overwrite the VP9 encoder's
15// internal reference frame. In the sample we set the last frame to the
16// current frame. This technique could be used to bounce between two cameras.
17//
18// The decoder would also have to set the reference frame to the same value
19// on the same frame, or the video will become corrupt. The 'test_decode'
20// variable is set to 1 in this example that tests if the encoder and decoder
21// results are matching.
22//
23// Usage
24// -----
25// This example encodes a raw video. And the last argument passed in specifies
26// the frame number to update the reference frame on. For example, run
27// examples/vp9cx_set_ref 352 288 in.yuv out.ivf 4 30
28// The parameter is parsed as follows:
29//
30//
31// Extra Variables
32// ---------------
33// This example maintains the frame number passed on the command line
34// in the `update_frame_num` variable.
35//
36//
37// Configuration
38// -------------
39//
40// The reference frame is updated on the frame specified on the command
41// line.
42//
43// Observing The Effects
44// ---------------------
45// The encoder and decoder results should be matching when the same reference
46// frame setting operation is done in both encoder and decoder. Otherwise,
47// the encoder/decoder mismatch would be seen.
48
49#include <stdio.h>
50#include <stdlib.h>
51#include <string.h>
52
53#include "vpx/vp8cx.h"
54#include "vpx/vpx_decoder.h"
55#include "vpx/vpx_encoder.h"
56#include "vp9/common/vp9_common.h"
57
58#include "./tools_common.h"
59#include "./video_writer.h"
60
61static const char *exec_name;
62
63void usage_exit() {
64 fprintf(stderr,
65 "Usage: %s <width> <height> <infile> <outfile> "
66 "<frame> <limit(optional)>\n",
67 exec_name);
68 exit(EXIT_FAILURE);
69}
70
71static int compare_img(const vpx_image_t *const img1,
72 const vpx_image_t *const img2) {
73 uint32_t l_w = img1->d_w;
74 uint32_t c_w = (img1->d_w + img1->x_chroma_shift) >> img1->x_chroma_shift;
75 const uint32_t c_h =
76 (img1->d_h + img1->y_chroma_shift) >> img1->y_chroma_shift;
77 uint32_t i;
78 int match = 1;
79
80 match &= (img1->fmt == img2->fmt);
81 match &= (img1->d_w == img2->d_w);
82 match &= (img1->d_h == img2->d_h);
83
84 for (i = 0; i < img1->d_h; ++i)
85 match &= (memcmp(img1->planes[VPX_PLANE_Y] + i * img1->stride[VPX_PLANE_Y],
86 img2->planes[VPX_PLANE_Y] + i * img2->stride[VPX_PLANE_Y],
87 l_w) == 0);
88
89 for (i = 0; i < c_h; ++i)
90 match &= (memcmp(img1->planes[VPX_PLANE_U] + i * img1->stride[VPX_PLANE_U],
91 img2->planes[VPX_PLANE_U] + i * img2->stride[VPX_PLANE_U],
92 c_w) == 0);
93
94 for (i = 0; i < c_h; ++i)
95 match &= (memcmp(img1->planes[VPX_PLANE_V] + i * img1->stride[VPX_PLANE_V],
96 img2->planes[VPX_PLANE_V] + i * img2->stride[VPX_PLANE_V],
97 c_w) == 0);
98
99 return match;
100}
101
102#define mmin(a, b) ((a) < (b) ? (a) : (b))
103static void find_mismatch(const vpx_image_t *const img1,
104 const vpx_image_t *const img2, int yloc[4],
105 int uloc[4], int vloc[4]) {
106 const uint32_t bsize = 64;
107 const uint32_t bsizey = bsize >> img1->y_chroma_shift;
108 const uint32_t bsizex = bsize >> img1->x_chroma_shift;
109 const uint32_t c_w =
110 (img1->d_w + img1->x_chroma_shift) >> img1->x_chroma_shift;
111 const uint32_t c_h =
112 (img1->d_h + img1->y_chroma_shift) >> img1->y_chroma_shift;
113 int match = 1;
114 uint32_t i, j;
115 yloc[0] = yloc[1] = yloc[2] = yloc[3] = -1;
116 for (i = 0, match = 1; match && i < img1->d_h; i += bsize) {
117 for (j = 0; match && j < img1->d_w; j += bsize) {
118 int k, l;
119 const int si = mmin(i + bsize, img1->d_h) - i;
120 const int sj = mmin(j + bsize, img1->d_w) - j;
121 for (k = 0; match && k < si; ++k) {
122 for (l = 0; match && l < sj; ++l) {
123 if (*(img1->planes[VPX_PLANE_Y] +
124 (i + k) * img1->stride[VPX_PLANE_Y] + j + l) !=
125 *(img2->planes[VPX_PLANE_Y] +
126 (i + k) * img2->stride[VPX_PLANE_Y] + j + l)) {
127 yloc[0] = i + k;
128 yloc[1] = j + l;
129 yloc[2] = *(img1->planes[VPX_PLANE_Y] +
130 (i + k) * img1->stride[VPX_PLANE_Y] + j + l);
131 yloc[3] = *(img2->planes[VPX_PLANE_Y] +
132 (i + k) * img2->stride[VPX_PLANE_Y] + j + l);
133 match = 0;
134 break;
135 }
136 }
137 }
138 }
139 }
140
141 uloc[0] = uloc[1] = uloc[2] = uloc[3] = -1;
142 for (i = 0, match = 1; match && i < c_h; i += bsizey) {
143 for (j = 0; match && j < c_w; j += bsizex) {
144 int k, l;
145 const int si = mmin(i + bsizey, c_h - i);
146 const int sj = mmin(j + bsizex, c_w - j);
147 for (k = 0; match && k < si; ++k) {
148 for (l = 0; match && l < sj; ++l) {
149 if (*(img1->planes[VPX_PLANE_U] +
150 (i + k) * img1->stride[VPX_PLANE_U] + j + l) !=
151 *(img2->planes[VPX_PLANE_U] +
152 (i + k) * img2->stride[VPX_PLANE_U] + j + l)) {
153 uloc[0] = i + k;
154 uloc[1] = j + l;
155 uloc[2] = *(img1->planes[VPX_PLANE_U] +
156 (i + k) * img1->stride[VPX_PLANE_U] + j + l);
157 uloc[3] = *(img2->planes[VPX_PLANE_U] +
158 (i + k) * img2->stride[VPX_PLANE_U] + j + l);
159 match = 0;
160 break;
161 }
162 }
163 }
164 }
165 }
166 vloc[0] = vloc[1] = vloc[2] = vloc[3] = -1;
167 for (i = 0, match = 1; match && i < c_h; i += bsizey) {
168 for (j = 0; match && j < c_w; j += bsizex) {
169 int k, l;
170 const int si = mmin(i + bsizey, c_h - i);
171 const int sj = mmin(j + bsizex, c_w - j);
172 for (k = 0; match && k < si; ++k) {
173 for (l = 0; match && l < sj; ++l) {
174 if (*(img1->planes[VPX_PLANE_V] +
175 (i + k) * img1->stride[VPX_PLANE_V] + j + l) !=
176 *(img2->planes[VPX_PLANE_V] +
177 (i + k) * img2->stride[VPX_PLANE_V] + j + l)) {
178 vloc[0] = i + k;
179 vloc[1] = j + l;
180 vloc[2] = *(img1->planes[VPX_PLANE_V] +
181 (i + k) * img1->stride[VPX_PLANE_V] + j + l);
182 vloc[3] = *(img2->planes[VPX_PLANE_V] +
183 (i + k) * img2->stride[VPX_PLANE_V] + j + l);
184 match = 0;
185 break;
186 }
187 }
188 }
189 }
190 }
191}
192
193static void testing_decode(vpx_codec_ctx_t *encoder, vpx_codec_ctx_t *decoder,
194 unsigned int frame_out, int *mismatch_seen) {
195 vpx_image_t enc_img, dec_img;
196 struct vp9_ref_frame ref_enc, ref_dec;
197
198 if (*mismatch_seen) return;
199
200 ref_enc.idx = 0;
201 ref_dec.idx = 0;
202 if (vpx_codec_control(encoder, VP9_GET_REFERENCE, &ref_enc))
203 die_codec(encoder, "Failed to get encoder reference frame");
204 enc_img = ref_enc.img;
205 if (vpx_codec_control(decoder, VP9_GET_REFERENCE, &ref_dec))
206 die_codec(decoder, "Failed to get decoder reference frame");
207 dec_img = ref_dec.img;
208
209 if (!compare_img(&enc_img, &dec_img)) {
210 int y[4], u[4], v[4];
211
212 *mismatch_seen = 1;
213
214 find_mismatch(&enc_img, &dec_img, y, u, v);
215 printf(
216 "Encode/decode mismatch on frame %d at"
217 " Y[%d, %d] {%d/%d},"
218 " U[%d, %d] {%d/%d},"
219 " V[%d, %d] {%d/%d}",
220 frame_out, y[0], y[1], y[2], y[3], u[0], u[1], u[2], u[3], v[0], v[1],
221 v[2], v[3]);
222 }
223
224 vpx_img_free(&enc_img);
225 vpx_img_free(&dec_img);
226}
227
228static int encode_frame(vpx_codec_ctx_t *ecodec, vpx_image_t *img,
229 unsigned int frame_in, VpxVideoWriter *writer,
230 int test_decode, vpx_codec_ctx_t *dcodec,
231 unsigned int *frame_out, int *mismatch_seen) {
232 int got_pkts = 0;
233 vpx_codec_iter_t iter = NULL;
234 const vpx_codec_cx_pkt_t *pkt = NULL;
235 int got_data;
236 const vpx_codec_err_t res =
237 vpx_codec_encode(ecodec, img, frame_in, 1, 0, VPX_DL_GOOD_QUALITY);
238 if (res != VPX_CODEC_OK) die_codec(ecodec, "Failed to encode frame");
239
240 got_data = 0;
241
242 while ((pkt = vpx_codec_get_cx_data(ecodec, &iter)) != NULL) {
243 got_pkts = 1;
244
245 if (pkt->kind == VPX_CODEC_CX_FRAME_PKT) {
246 const int keyframe = (pkt->data.frame.flags & VPX_FRAME_IS_KEY) != 0;
247
248 if (!(pkt->data.frame.flags & VPX_FRAME_IS_FRAGMENT)) {
249 *frame_out += 1;
250 }
251
252 if (!vpx_video_writer_write_frame(writer, pkt->data.frame.buf,
253 pkt->data.frame.sz,
254 pkt->data.frame.pts)) {
255 die_codec(ecodec, "Failed to write compressed frame");
256 }
257 printf(keyframe ? "K" : ".");
258 fflush(stdout);
259 got_data = 1;
260
261 // Decode 1 frame.
262 if (test_decode) {
263 if (vpx_codec_decode(dcodec, pkt->data.frame.buf,
264 (unsigned int)pkt->data.frame.sz, NULL, 0))
265 die_codec(dcodec, "Failed to decode frame.");
266 }
267 }
268 }
269
270 // Mismatch checking
271 if (got_data && test_decode) {
272 testing_decode(ecodec, dcodec, *frame_out, mismatch_seen);
273 }
274
275 return got_pkts;
276}
277
278int main(int argc, char **argv) {
279 FILE *infile = NULL;
280 // Encoder
281 vpx_codec_ctx_t ecodec;
283 unsigned int frame_in = 0;
284 vpx_image_t raw;
285 vpx_codec_err_t res;
286 VpxVideoInfo info;
287 VpxVideoWriter *writer = NULL;
288 const VpxInterface *encoder = NULL;
289
290 // Test encoder/decoder mismatch.
291 int test_decode = 1;
292 // Decoder
293 vpx_codec_ctx_t dcodec;
294 unsigned int frame_out = 0;
295
296 // The frame number to set reference frame on
297 unsigned int update_frame_num = 0;
298 int mismatch_seen = 0;
299
300 const int fps = 30;
301 const int bitrate = 500;
302
303 const char *width_arg = NULL;
304 const char *height_arg = NULL;
305 const char *infile_arg = NULL;
306 const char *outfile_arg = NULL;
307 const char *update_frame_num_arg = NULL;
308 unsigned int limit = 0;
309
310 vp9_zero(ecodec);
311 vp9_zero(cfg);
312 vp9_zero(info);
313
314 exec_name = argv[0];
315
316 if (argc < 6) die("Invalid number of arguments");
317
318 width_arg = argv[1];
319 height_arg = argv[2];
320 infile_arg = argv[3];
321 outfile_arg = argv[4];
322 update_frame_num_arg = argv[5];
323
324 encoder = get_vpx_encoder_by_name("vp9");
325 if (!encoder) die("Unsupported codec.");
326
327 update_frame_num = (unsigned int)strtoul(update_frame_num_arg, NULL, 0);
328 // In VP9, the reference buffers (cm->buffer_pool->frame_bufs[i].buf) are
329 // allocated while calling vpx_codec_encode(), thus, setting reference for
330 // 1st frame isn't supported.
331 if (update_frame_num <= 1) {
332 die("Couldn't parse frame number '%s'\n", update_frame_num_arg);
333 }
334
335 if (argc > 6) {
336 limit = (unsigned int)strtoul(argv[6], NULL, 0);
337 if (update_frame_num > limit)
338 die("Update frame number couldn't larger than limit\n");
339 }
340
341 info.codec_fourcc = encoder->fourcc;
342 info.frame_width = (int)strtol(width_arg, NULL, 0);
343 info.frame_height = (int)strtol(height_arg, NULL, 0);
344 info.time_base.numerator = 1;
345 info.time_base.denominator = fps;
346
347 if (info.frame_width <= 0 || info.frame_height <= 0 ||
348 (info.frame_width % 2) != 0 || (info.frame_height % 2) != 0) {
349 die("Invalid frame size: %dx%d", info.frame_width, info.frame_height);
350 }
351
352 if (!vpx_img_alloc(&raw, VPX_IMG_FMT_I420, info.frame_width,
353 info.frame_height, 1)) {
354 die("Failed to allocate image.");
355 }
356
357 printf("Using %s\n", vpx_codec_iface_name(encoder->codec_interface()));
358
359 res = vpx_codec_enc_config_default(encoder->codec_interface(), &cfg, 0);
360 if (res) die_codec(&ecodec, "Failed to get default codec config.");
361
362 cfg.g_w = info.frame_width;
363 cfg.g_h = info.frame_height;
364 cfg.g_timebase.num = info.time_base.numerator;
365 cfg.g_timebase.den = info.time_base.denominator;
366 cfg.rc_target_bitrate = bitrate;
367 cfg.g_lag_in_frames = 3;
368
369 writer = vpx_video_writer_open(outfile_arg, kContainerIVF, &info);
370 if (!writer) die("Failed to open %s for writing.", outfile_arg);
371
372 if (!(infile = fopen(infile_arg, "rb")))
373 die("Failed to open %s for reading.", infile_arg);
374
375 if (vpx_codec_enc_init(&ecodec, encoder->codec_interface(), &cfg, 0))
376 die_codec(&ecodec, "Failed to initialize encoder");
377
378 // Disable alt_ref.
380 die_codec(&ecodec, "Failed to set enable auto alt ref");
381
382 if (test_decode) {
383 const VpxInterface *decoder = get_vpx_decoder_by_name("vp9");
384 if (vpx_codec_dec_init(&dcodec, decoder->codec_interface(), NULL, 0))
385 die_codec(&dcodec, "Failed to initialize decoder.");
386 }
387
388 // Encode frames.
389 while (vpx_img_read(&raw, infile)) {
390 if (limit && frame_in >= limit) break;
391 if (update_frame_num > 1 && frame_out + 1 == update_frame_num) {
392 vpx_ref_frame_t ref;
393 ref.frame_type = VP8_LAST_FRAME;
394 ref.img = raw;
395 // Set reference frame in encoder.
396 if (vpx_codec_control(&ecodec, VP8_SET_REFERENCE, &ref))
397 die_codec(&ecodec, "Failed to set reference frame");
398 printf(" <SET_REF>");
399
400 // If set_reference in decoder is commented out, the enc/dec mismatch
401 // would be seen.
402 if (test_decode) {
403 if (vpx_codec_control(&dcodec, VP8_SET_REFERENCE, &ref))
404 die_codec(&dcodec, "Failed to set reference frame");
405 }
406 }
407
408 encode_frame(&ecodec, &raw, frame_in, writer, test_decode, &dcodec,
409 &frame_out, &mismatch_seen);
410 frame_in++;
411 if (mismatch_seen) break;
412 }
413
414 // Flush encoder.
415 if (!mismatch_seen)
416 while (encode_frame(&ecodec, NULL, frame_in, writer, test_decode, &dcodec,
417 &frame_out, &mismatch_seen)) {
418 }
419
420 printf("\n");
421 fclose(infile);
422 printf("Processed %d frames.\n", frame_out);
423
424 if (test_decode) {
425 if (!mismatch_seen)
426 printf("Encoder/decoder results are matching.\n");
427 else
428 printf("Encoder/decoder results are NOT matching.\n");
429 }
430
431 if (test_decode)
432 if (vpx_codec_destroy(&dcodec))
433 die_codec(&dcodec, "Failed to destroy decoder");
434
435 vpx_img_free(&raw);
436 if (vpx_codec_destroy(&ecodec))
437 die_codec(&ecodec, "Failed to destroy encoder.");
438
439 vpx_video_writer_close(writer);
440
441 return EXIT_SUCCESS;
442}
vpx_codec_err_t vpx_codec_destroy(vpx_codec_ctx_t *ctx)
Destroy a codec instance.
const void * vpx_codec_iter_t
Iterator.
Definition: vpx_codec.h:187
const char * vpx_codec_iface_name(vpx_codec_iface_t *iface)
Return the name for a given interface.
#define vpx_codec_control(ctx, id, data)
vpx_codec_control wrapper macro
Definition: vpx_codec.h:404
vpx_codec_err_t
Algorithm return codes.
Definition: vpx_codec.h:90
@ VPX_CODEC_OK
Operation completed without error.
Definition: vpx_codec.h:92
vpx_codec_err_t vpx_codec_decode(vpx_codec_ctx_t *ctx, const uint8_t *data, unsigned int data_sz, void *user_priv, long deadline)
Decode data.
#define vpx_codec_dec_init(ctx, iface, cfg, flags)
Convenience macro for vpx_codec_dec_init_ver()
Definition: vpx_decoder.h:144
#define vpx_codec_enc_init(ctx, iface, cfg, flags)
Convenience macro for vpx_codec_enc_init_ver()
Definition: vpx_encoder.h:760
#define VPX_DL_GOOD_QUALITY
deadline parameter analogous to VPx GOOD QUALITY mode.
Definition: vpx_encoder.h:851
const vpx_codec_cx_pkt_t * vpx_codec_get_cx_data(vpx_codec_ctx_t *ctx, vpx_codec_iter_t *iter)
Encoded data iterator.
vpx_codec_err_t vpx_codec_enc_config_default(vpx_codec_iface_t *iface, vpx_codec_enc_cfg_t *cfg, unsigned int reserved)
Get a default configuration.
#define VPX_FRAME_IS_KEY
Definition: vpx_encoder.h:122
vpx_codec_err_t vpx_codec_encode(vpx_codec_ctx_t *ctx, const vpx_image_t *img, vpx_codec_pts_t pts, unsigned long duration, vpx_enc_frame_flags_t flags, unsigned long deadline)
Encode a frame.
#define VPX_FRAME_IS_FRAGMENT
this is a fragment of the encoded frame
Definition: vpx_encoder.h:129
@ VPX_CODEC_CX_FRAME_PKT
Definition: vpx_encoder.h:153
@ VP8E_SET_ENABLEAUTOALTREF
Codec control function to enable automatic set and use alf frames.
Definition: vp8cx.h:161
@ VP8_SET_REFERENCE
pass in an external frame into decoder to be used as reference frame
Definition: vp8.h:47
@ VP9_GET_REFERENCE
Definition: vp8.h:59
VP9 specific reference frame data struct.
Definition: vp8.h:119
int idx
Definition: vp8.h:120
vpx_image_t img
Definition: vp8.h:121
Codec context structure.
Definition: vpx_codec.h:197
Encoder output packet.
Definition: vpx_encoder.h:170
vpx_codec_frame_flags_t flags
Definition: vpx_encoder.h:180
enum vpx_codec_cx_pkt_kind kind
Definition: vpx_encoder.h:171
struct vpx_codec_cx_pkt::@1::@2 frame
size_t sz
Definition: vpx_encoder.h:175
void * buf
Definition: vpx_encoder.h:174
vpx_codec_pts_t pts
time stamp to show frame (in timebase units)
Definition: vpx_encoder.h:177
union vpx_codec_cx_pkt::@1 data
Encoder configuration structure.
Definition: vpx_encoder.h:279
unsigned int g_h
Height of the frame.
Definition: vpx_encoder.h:327
unsigned int g_w
Width of the frame.
Definition: vpx_encoder.h:318
struct vpx_rational g_timebase
Stream timebase units.
Definition: vpx_encoder.h:357
unsigned int g_lag_in_frames
Allow lagged encoding.
Definition: vpx_encoder.h:386
unsigned int rc_target_bitrate
Target data rate.
Definition: vpx_encoder.h:477
Image Descriptor.
Definition: vpx_image.h:88
vpx_img_fmt_t fmt
Definition: vpx_image.h:89
unsigned int y_chroma_shift
Definition: vpx_image.h:108
unsigned int d_h
Definition: vpx_image.h:100
unsigned int d_w
Definition: vpx_image.h:99
unsigned char * planes[4]
Definition: vpx_image.h:116
int stride[4]
Definition: vpx_image.h:117
unsigned int x_chroma_shift
Definition: vpx_image.h:107
int den
Definition: vpx_encoder.h:231
int num
Definition: vpx_encoder.h:230
reference frame data struct
Definition: vp8.h:110
vpx_ref_frame_type_t frame_type
Definition: vp8.h:111
vpx_image_t img
Definition: vp8.h:112
Provides definitions for using VP8 or VP9 encoder algorithm within the vpx Codec Interface.
Describes the decoder algorithm interface to applications.
Describes the encoder algorithm interface to applications.
#define VPX_PLANE_Y
Definition: vpx_image.h:112
vpx_image_t * vpx_img_alloc(vpx_image_t *img, vpx_img_fmt_t fmt, unsigned int d_w, unsigned int d_h, unsigned int align)
Open a descriptor, allocating storage for the underlying image.
#define VPX_PLANE_U
Definition: vpx_image.h:113
@ VPX_IMG_FMT_I420
Definition: vpx_image.h:55
#define VPX_PLANE_V
Definition: vpx_image.h:114
void vpx_img_free(vpx_image_t *img)
Close an image descriptor.