AOMedia AV1 Codec
lightfield_bitstream_parsing
1 /*
2  * Copyright (c) 2018, Alliance for Open Media. All rights reserved.
3  *
4  * This source code is subject to the terms of the BSD 2 Clause License and
5  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6  * was not distributed with this source code in the LICENSE file, you can
7  * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8  * Media Patent License 1.0 was not distributed with this source code in the
9  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10  */
11 
12 // Lightfield Bitstream Parsing
13 // ============================
14 //
15 // This is a lightfield bitstream parsing example. It takes an input file
16 // containing the whole compressed lightfield bitstream(ivf file) and a text
17 // file containing a stream of tiles to decode and then constructs and outputs
18 // a new bitstream that can be decoded by an AV1 decoder. The output bitstream
19 // contains reference frames(i.e. anchor frames), camera frame header, and
20 // tile list OBUs. num_references is the number of anchor frames coded at the
21 // beginning of the light field file. After running the lightfield encoder,
22 // run lightfield bitstream parsing:
23 // examples/lightfield_bitstream_parsing vase10x10.ivf vase_tile_list.ivf 4
24 // tile_list.txt
25 //
26 // The tile_list.txt is expected to be of the form:
27 // Frame <frame_index0>
28 // <image_index0> <anchor_index0> <tile_col0> <tile_row0>
29 // <image_index1> <anchor_index1> <tile_col1> <tile_row1>
30 // ...
31 // Frame <frame_index1)
32 // ...
33 //
34 // The "Frame" markers indicate a new render frame and thus a new tile list
35 // will be started and the old one flushed. The image_indexN, anchor_indexN,
36 // tile_colN, and tile_rowN identify an individual tile to be decoded and
37 // to use anchor_indexN anchor image for MCP.
38 
39 #include <stdio.h>
40 #include <stdlib.h>
41 #include <string.h>
42 
43 #include "aom/aom_codec.h"
44 #include "aom/aom_decoder.h"
45 #include "aom/aom_encoder.h"
46 #include "aom/aom_integer.h"
47 #include "aom/aomdx.h"
48 #include "common/tools_common.h"
49 #include "common/video_reader.h"
50 #include "common/video_writer.h"
51 
52 #define MAX_TILES 512
53 
54 static const char *exec_name;
55 
56 void usage_exit(void) {
57  fprintf(stderr, "Usage: %s <infile> <outfile> <num_references> <tile_list>\n",
58  exec_name);
59  exit(EXIT_FAILURE);
60 }
61 
62 #define ALIGN_POWER_OF_TWO(value, n) \
63  (((value) + ((1 << (n)) - 1)) & ~((1 << (n)) - 1))
64 
65 const int output_frame_width = 512;
66 const int output_frame_height = 512;
67 
68 // Spec:
69 // typedef struct {
70 // uint8_t anchor_frame_idx;
71 // uint8_t tile_row;
72 // uint8_t tile_col;
73 // uint16_t coded_tile_data_size_minus_1;
74 // uint8_t *coded_tile_data;
75 // } TILE_LIST_ENTRY;
76 
77 // Tile list entry provided by the application
78 typedef struct {
79  int image_idx;
80  int reference_idx;
81  int tile_col;
82  int tile_row;
83 } TILE_LIST_INFO;
84 
85 static int get_image_bps(aom_img_fmt_t fmt) {
86  switch (fmt) {
87  case AOM_IMG_FMT_I420: return 12;
88  case AOM_IMG_FMT_I422: return 16;
89  case AOM_IMG_FMT_I444: return 24;
90  case AOM_IMG_FMT_I42016: return 24;
91  case AOM_IMG_FMT_I42216: return 32;
92  case AOM_IMG_FMT_I44416: return 48;
93  default: die("Invalid image format");
94  }
95 }
96 
97 static void process_tile_list(const TILE_LIST_INFO *tiles, int num_tiles,
98  aom_codec_pts_t tl_pts, unsigned char **frames,
99  const size_t *frame_sizes, aom_codec_ctx_t *codec,
100  unsigned char *tl_buf, AvxVideoWriter *writer,
101  uint8_t output_frame_width_in_tiles_minus_1,
102  uint8_t output_frame_height_in_tiles_minus_1) {
103  unsigned char *tl = tl_buf;
104  unsigned char *saved_obu_size_loc = NULL;
105  uint32_t tile_list_obu_header_size = 0;
106  uint32_t tile_list_obu_size = 0;
107  int num_tiles_minus_1 = num_tiles - 1;
108  int i;
109 
110  // Write the tile list OBU header that is 1 byte long.
111  int obu_type = OBU_TILE_LIST;
112  int obu_has_size_field = 1;
113  *tl++ = (obu_type << 3) | (obu_has_size_field << 1);
114  tile_list_obu_header_size++;
115 
116  // Write the OBU size using a fixed length_field_size of 4 bytes.
117  saved_obu_size_loc = tl;
118  for (i = 0; i < 4; i++) {
119  *tl++ = 0;
120  }
121  tile_list_obu_header_size += 4;
122 
123  // write_tile_list_obu()
124  *tl++ = output_frame_width_in_tiles_minus_1;
125  *tl++ = output_frame_height_in_tiles_minus_1;
126  *tl++ = (num_tiles_minus_1 >> 8) & 0xff;
127  *tl++ = num_tiles_minus_1 & 0xff;
128  tile_list_obu_size += 4;
129 
130  // Write each tile's data
131  for (i = 0; i <= num_tiles_minus_1; i++) {
132  aom_tile_data tile_data = { 0, NULL, 0 };
133 
134  int image_idx = tiles[i].image_idx;
135  int ref_idx = tiles[i].reference_idx;
136  int tc = tiles[i].tile_col;
137  int tr = tiles[i].tile_row;
138 
139  size_t frame_size = frame_sizes[image_idx];
140  const unsigned char *frame = frames[image_idx];
141 
143  AOM_CODEC_CONTROL_TYPECHECKED(codec, AV1_SET_DECODE_TILE_COL, tc);
144 
145  aom_codec_err_t aom_status =
146  aom_codec_decode(codec, frame, frame_size, NULL);
147  if (aom_status) die_codec(codec, "Failed to decode tile.");
148 
150 
151  // Copy over tile info.
152  // uint8_t anchor_frame_idx;
153  // uint8_t tile_row;
154  // uint8_t tile_col;
155  // uint16_t coded_tile_data_size_minus_1;
156  // uint8_t *coded_tile_data;
157  uint32_t tile_info_bytes = 5;
158  *tl++ = ref_idx;
159  *tl++ = tr;
160  *tl++ = tc;
161  int coded_tile_data_size_minus_1 = (int)tile_data.coded_tile_data_size - 1;
162  *tl++ = (coded_tile_data_size_minus_1 >> 8) & 0xff;
163  *tl++ = coded_tile_data_size_minus_1 & 0xff;
164 
165  memcpy(tl, (uint8_t *)tile_data.coded_tile_data,
166  tile_data.coded_tile_data_size);
167  tl += tile_data.coded_tile_data_size;
168 
169  tile_list_obu_size +=
170  tile_info_bytes + (uint32_t)tile_data.coded_tile_data_size;
171  }
172 
173  // Write tile list OBU size.
174  size_t bytes_written = 0;
175  if (aom_uleb_encode_fixed_size(tile_list_obu_size, 4, 4, saved_obu_size_loc,
176  &bytes_written))
177  die_codec(codec, "Failed to encode the tile list obu size.");
178 
179  // Copy the tile list.
180  if (!aom_video_writer_write_frame(
181  writer, tl_buf, tile_list_obu_header_size + tile_list_obu_size,
182  tl_pts))
183  die_codec(codec, "Failed to copy compressed tile list.");
184 }
185 
186 int main(int argc, char **argv) {
187  AvxVideoReader *reader = NULL;
188  AvxVideoWriter *writer = NULL;
189  const AvxVideoInfo *info = NULL;
190  int num_references;
191  int i;
192  aom_codec_pts_t pts;
193  const char *tile_list_file = NULL;
194 
195  exec_name = argv[0];
196  if (argc != 5) die("Invalid number of arguments.");
197 
198  reader = aom_video_reader_open(argv[1]);
199  if (!reader) die("Failed to open %s for reading.", argv[1]);
200 
201  num_references = (int)strtol(argv[3], NULL, 0);
202  info = aom_video_reader_get_info(reader);
203 
204  aom_video_reader_set_fourcc(reader, AV1_FOURCC);
205 
206  // The writer to write out ivf file in tile list OBU, which can be decoded by
207  // AV1 decoder.
208  writer = aom_video_writer_open(argv[2], kContainerIVF, info);
209  if (!writer) die("Failed to open %s for writing", argv[2]);
210 
211  tile_list_file = argv[4];
212 
213  aom_codec_iface_t *decoder = get_aom_decoder_by_fourcc(info->codec_fourcc);
214  if (!decoder) die("Unknown input codec.");
215  printf("Using %s\n", aom_codec_iface_name(decoder));
216 
217  aom_codec_ctx_t codec;
218  if (aom_codec_dec_init(&codec, decoder, NULL, 0))
219  die("Failed to initialize decoder.");
220 
221  // Decode anchor frames.
223 
224  printf("Reading %d reference images.\n", num_references);
225  for (i = 0; i < num_references; ++i) {
226  aom_video_reader_read_frame(reader);
227 
228  size_t frame_size = 0;
229  const unsigned char *frame =
230  aom_video_reader_get_frame(reader, &frame_size);
231  pts = (aom_codec_pts_t)aom_video_reader_get_frame_pts(reader);
232 
233  // Copy references bitstream directly.
234  if (!aom_video_writer_write_frame(writer, frame, frame_size, pts))
235  die_codec(&codec, "Failed to copy compressed anchor frame.");
236 
237  if (aom_codec_decode(&codec, frame, frame_size, NULL))
238  die_codec(&codec, "Failed to decode frame.");
239  }
240 
241  // Decode camera frames.
244 
245  FILE *infile = aom_video_reader_get_file(reader);
246  // Record the offset of the first camera image.
247  const FileOffset camera_frame_pos = ftello(infile);
248 
249  printf("Loading compressed frames into memory.\n");
250 
251  // Count the frames in the lightfield.
252  int num_frames = 0;
253  while (aom_video_reader_read_frame(reader)) {
254  ++num_frames;
255  }
256  if (num_frames < 1) die("Input light field has no frames.");
257 
258  // Read all of the lightfield frames into memory.
259  unsigned char **frames =
260  (unsigned char **)malloc(num_frames * sizeof(unsigned char *));
261  size_t *frame_sizes = (size_t *)malloc(num_frames * sizeof(size_t));
262  if (!(frames && frame_sizes)) die("Failed to allocate frame data.");
263 
264  // Seek to the first camera image.
265  fseeko(infile, camera_frame_pos, SEEK_SET);
266  for (int f = 0; f < num_frames; ++f) {
267  aom_video_reader_read_frame(reader);
268  size_t frame_size = 0;
269  const unsigned char *frame =
270  aom_video_reader_get_frame(reader, &frame_size);
271  frames[f] = (unsigned char *)malloc(frame_size * sizeof(unsigned char));
272  if (!frames[f]) die("Failed to allocate frame data.");
273  memcpy(frames[f], frame, frame_size);
274  frame_sizes[f] = frame_size;
275  }
276  printf("Read %d frames.\n", num_frames);
277 
278  // Copy first camera frame for getting camera frame header. This is done
279  // only once.
280  {
281  size_t frame_size = frame_sizes[0];
282  const unsigned char *frame = frames[0];
283  pts = num_references;
284  aom_tile_data frame_header_info = { 0, NULL, 0 };
285 
286  // Need to decode frame header to get camera frame header info. So, here
287  // decoding 1 tile is enough.
289  AOM_CODEC_CONTROL_TYPECHECKED(&codec, AV1_SET_DECODE_TILE_COL, 0);
290 
291  aom_codec_err_t aom_status =
292  aom_codec_decode(&codec, frame, frame_size, NULL);
293  if (aom_status) die_codec(&codec, "Failed to decode tile.");
294 
296  &frame_header_info);
297 
298  size_t obu_size_offset =
299  (uint8_t *)frame_header_info.coded_tile_data - frame;
300  size_t length_field_size = frame_header_info.coded_tile_data_size;
301  // Remove ext-tile tile info.
302  uint32_t frame_header_size = (uint32_t)frame_header_info.extra_size - 1;
303  size_t bytes_to_copy =
304  obu_size_offset + length_field_size + frame_header_size;
305 
306  unsigned char *frame_hdr_buf = (unsigned char *)malloc(bytes_to_copy);
307  if (frame_hdr_buf == NULL)
308  die_codec(&codec, "Failed to allocate frame header buffer.");
309 
310  memcpy(frame_hdr_buf, frame, bytes_to_copy);
311 
312  // Update frame header OBU size.
313  size_t bytes_written = 0;
314  if (aom_uleb_encode_fixed_size(
315  frame_header_size, length_field_size, length_field_size,
316  frame_hdr_buf + obu_size_offset, &bytes_written))
317  die_codec(&codec, "Failed to encode the tile list obu size.");
318 
319  // Copy camera frame header bitstream.
320  if (!aom_video_writer_write_frame(writer, frame_hdr_buf, bytes_to_copy,
321  pts))
322  die_codec(&codec, "Failed to copy compressed camera frame header.");
323  free(frame_hdr_buf);
324  }
325 
326  // Read out the image format.
327  aom_img_fmt_t ref_fmt = 0;
328  if (AOM_CODEC_CONTROL_TYPECHECKED(&codec, AV1D_GET_IMG_FORMAT, &ref_fmt))
329  die_codec(&codec, "Failed to get the image format");
330  const int bps = get_image_bps(ref_fmt);
331  if (!bps) die_codec(&codec, "Invalid image format.");
332  // read out the tile size.
333  unsigned int tile_size = 0;
334  if (AOM_CODEC_CONTROL_TYPECHECKED(&codec, AV1D_GET_TILE_SIZE, &tile_size))
335  die_codec(&codec, "Failed to get the tile size");
336  const unsigned int tile_width = tile_size >> 16;
337  const unsigned int tile_height = tile_size & 65535;
338  // Allocate a buffer to store tile list bitstream.
339  const size_t data_sz = MAX_TILES * ALIGN_POWER_OF_TWO(tile_width, 5) *
340  ALIGN_POWER_OF_TWO(tile_height, 5) * bps / 8;
341 
342  unsigned char *tl_buf = (unsigned char *)malloc(data_sz);
343  if (tl_buf == NULL) die_codec(&codec, "Failed to allocate tile list buffer.");
344 
345  aom_codec_pts_t tl_pts = num_references;
346  const uint8_t output_frame_width_in_tiles_minus_1 =
347  output_frame_width / tile_width - 1;
348  const uint8_t output_frame_height_in_tiles_minus_1 =
349  output_frame_height / tile_height - 1;
350 
351  printf("Reading tile list from file.\n");
352  char line[1024];
353  FILE *tile_list_fptr = fopen(tile_list_file, "r");
354  if (!tile_list_fptr) die_codec(&codec, "Failed to open tile list file.");
355  int num_tiles = 0;
356  TILE_LIST_INFO tiles[MAX_TILES];
357  while ((fgets(line, 1024, tile_list_fptr)) != NULL) {
358  if (line[0] == 'F' || num_tiles >= MAX_TILES) {
359  // Flush existing tile list and start another, either because we hit a
360  // new render frame or because we've hit our max number of tiles per list.
361  if (num_tiles > 0) {
362  process_tile_list(tiles, num_tiles, tl_pts, frames, frame_sizes, &codec,
363  tl_buf, writer, output_frame_width_in_tiles_minus_1,
364  output_frame_height_in_tiles_minus_1);
365  ++tl_pts;
366  }
367  num_tiles = 0;
368  }
369  if (line[0] == 'F') {
370  continue;
371  }
372  if (sscanf(line, "%d %d %d %d", &tiles[num_tiles].image_idx,
373  &tiles[num_tiles].reference_idx, &tiles[num_tiles].tile_col,
374  &tiles[num_tiles].tile_row) == 4) {
375  if (tiles[num_tiles].image_idx >= num_frames) {
376  die("Tile list image_idx out of bounds: %d >= %d.",
377  tiles[num_tiles].image_idx, num_frames);
378  }
379  if (tiles[num_tiles].reference_idx >= num_references) {
380  die("Tile list reference_idx out of bounds: %d >= %d.",
381  tiles[num_tiles].reference_idx, num_references);
382  }
383  ++num_tiles;
384  }
385  }
386  if (num_tiles > 0) {
387  // Flush out the last tile list.
388  process_tile_list(tiles, num_tiles, tl_pts, frames, frame_sizes, &codec,
389  tl_buf, writer, output_frame_width_in_tiles_minus_1,
390  output_frame_height_in_tiles_minus_1);
391  ++tl_pts;
392  }
393 
394  const int num_tile_lists = (int)(tl_pts - pts);
395  printf("Finished processing tile lists. Num tile lists: %d.\n",
396  num_tile_lists);
397  free(tl_buf);
398  for (int f = 0; f < num_frames; ++f) {
399  free(frames[f]);
400  }
401  free(frame_sizes);
402  free(frames);
403  if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec");
404  aom_video_writer_close(writer);
405  aom_video_reader_close(reader);
406 
407  return EXIT_SUCCESS;
408 }
#define AOM_CODEC_CONTROL_TYPECHECKED(ctx, id, data)
aom_codec_control wrapper macro (adds type-checking, less flexible)
Definition: aom_codec.h:542
Codec control function to get the frame header information of an encoded frame, aom_tile_data* parame...
Definition: aomdx.h:321
int64_t aom_codec_pts_t
Time Stamp Type.
Definition: aom_codec.h:252
Describes the encoder algorithm interface to applications.
enum aom_img_fmt aom_img_fmt_t
List of supported image formats.
Codec control function to get the start address and size of a tile in the coded bitstream, aom_tile_data* parameter.
Definition: aomdx.h:326
Definition: aom_image.h:50
Codec context structure.
Definition: aom_codec.h:315
Codec control function to get the width and height (in pixels) of the tiles in a tile list...
Definition: aomdx.h:243
Describes the decoder algorithm interface to applications.
Definition: aom_image.h:49
Structure to hold a tile's start address and size in the bitstream.
Definition: aomdx.h:91
Definition: aom_image.h:59
aom_codec_err_t aom_codec_decode(aom_codec_ctx_t *ctx, const uint8_t *data, size_t data_sz, void *user_priv)
Decode data.
const struct aom_codec_iface aom_codec_iface_t
Codec interface structure.
Definition: aom_codec.h:271
#define aom_codec_dec_init(ctx, iface, cfg, flags)
Convenience macro for aom_codec_dec_init_ver()
Definition: aom_decoder.h:129
const char * aom_codec_iface_name(aom_codec_iface_t *iface)
Return the name for a given interface.
aom_codec_err_t aom_codec_destroy(aom_codec_ctx_t *ctx)
Destroy a codec instance.
Definition: aom_image.h:58
Codec control function to get the image format of the stream, aom_img_fmt_t* parameter.
Definition: aomdx.h:235
Describes the codec algorithm interface to applications.
Codec control function to enable the ext-tile software debug and testing code in the decoder...
Definition: aomdx.h:339
size_t extra_size
Definition: aomdx.h:97
Codec control function to set the range of tile decoding, int parameter.
Definition: aomdx.h:307
size_t coded_tile_data_size
Definition: aomdx.h:93
Codec control function to set the tile coding mode, unsigned int parameter.
Definition: aomdx.h:316
const void * coded_tile_data
Definition: aomdx.h:95
aom_codec_err_t
Algorithm return codes.
Definition: aom_codec.h:155
Definition: aom_image.h:56
Provides definitions for using AOM or AV1 within the aom Decoder interface.
Definition: aom_image.h:45