// TODO(tomfinegan): Improve command line parsing and add args for bitrate/fps. int main(int argc, char **argv) { FILE *infile = NULL; aom_codec_ctx_t codec; aom_codec_enc_cfg_t cfg; int frame_count = 0; aom_image_t raw; aom_codec_err_t res; AvxVideoInfo info; AvxVideoWriter *writer = NULL; const AvxInterface *encoder = NULL; const int fps = 30; const int bitrate = 200; int keyframe_interval = 0; int max_frames = 0; int frames_encoded = 0; const char *codec_arg = NULL; const char *width_arg = NULL; const char *height_arg = NULL; const char *infile_arg = NULL; const char *outfile_arg = NULL; const char *keyframe_interval_arg = NULL; exec_name = argv[0]; // Clear explicitly, as simply assigning "{ 0 }" generates // "missing-field-initializers" warning in some compilers. memset(&info, 0, sizeof(info)); if (argc != 9) die("Invalid number of arguments"); codec_arg = argv[1]; width_arg = argv[2]; height_arg = argv[3]; infile_arg = argv[4]; outfile_arg = argv[5]; keyframe_interval_arg = argv[6]; max_frames = (int)strtol(argv[8], NULL, 0); encoder = get_aom_encoder_by_name(codec_arg); if (!encoder) die("Unsupported codec."); info.codec_fourcc = encoder->fourcc; info.frame_width = (int)strtol(width_arg, NULL, 0); info.frame_height = (int)strtol(height_arg, NULL, 0); info.time_base.numerator = 1; info.time_base.denominator = fps; if (info.frame_width <= 0 || info.frame_height <= 0 || (info.frame_width % 2) != 0 || (info.frame_height % 2) != 0) { die("Invalid frame size: %dx%d", info.frame_width, info.frame_height); } if (!aom_img_alloc(&raw, AOM_IMG_FMT_I420, info.frame_width, info.frame_height, 1)) { die("Failed to allocate image."); } keyframe_interval = (int)strtol(keyframe_interval_arg, NULL, 0); if (keyframe_interval < 0) die("Invalid keyframe interval value."); printf("Using %s\n", aom_codec_iface_name(encoder->codec_interface())); res = aom_codec_enc_config_default(encoder->codec_interface(), &cfg, 0); if (res) die_codec(&codec, "Failed to get default codec config."); cfg.g_w = info.frame_width; cfg.g_h = info.frame_height; cfg.g_timebase.num = info.time_base.numerator; cfg.g_timebase.den = info.time_base.denominator; cfg.rc_target_bitrate = bitrate; cfg.g_error_resilient = (aom_codec_er_flags_t)strtoul(argv[7], NULL, 0); writer = aom_video_writer_open(outfile_arg, kContainerIVF, &info); if (!writer) die("Failed to open %s for writing.", outfile_arg); if (!(infile = fopen(infile_arg, "rb"))) die("Failed to open %s for reading.", infile_arg); if (aom_codec_enc_init(&codec, encoder->codec_interface(), &cfg, 0)) die_codec(&codec, "Failed to initialize encoder"); // Encode frames. while (aom_img_read(&raw, infile)) { int flags = 0; if (keyframe_interval > 0 && frame_count % keyframe_interval == 0) flags |= AOM_EFLAG_FORCE_KF; encode_frame(&codec, &raw, frame_count++, flags, writer); frames_encoded++; if (max_frames > 0 && frames_encoded >= max_frames) break; } // Flush encoder. while (encode_frame(&codec, NULL, -1, 0, writer)) continue; printf("\n"); fclose(infile); printf("Processed %d frames.\n", frame_count); aom_img_free(&raw); if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec."); aom_video_writer_close(writer); return EXIT_SUCCESS; }
static void pass1(aom_image_t *raw, FILE *infile, const char *outfile_name, const AvxInterface *encoder, aom_codec_enc_cfg_t *cfg, int lf_width, int lf_height, int lf_blocksize, int flags, aom_image_t *raw_shift) { AvxVideoInfo info = { encoder->fourcc, cfg->g_w, cfg->g_h, { cfg->g_timebase.num, cfg->g_timebase.den }, 0 }; AvxVideoWriter *writer = NULL; aom_codec_ctx_t codec; int frame_count = 0; int image_size_bytes = aom_img_size_bytes(raw); int bu, bv; int u_blocks, v_blocks; aom_image_t *frame_to_encode; aom_image_t reference_images[MAX_EXTERNAL_REFERENCES]; int reference_image_num = 0; int i; writer = aom_video_writer_open(outfile_name, kContainerIVF, &info); if (!writer) die("Failed to open %s for writing", outfile_name); if (aom_codec_enc_init(&codec, encoder->codec_interface(), cfg, flags)) die_codec(&codec, "Failed to initialize encoder"); if (aom_codec_control(&codec, AOME_SET_ENABLEAUTOALTREF, 0)) die_codec(&codec, "Failed to turn off auto altref"); if (aom_codec_control(&codec, AV1E_SET_FRAME_PARALLEL_DECODING, 0)) die_codec(&codec, "Failed to set frame parallel decoding"); // Note: The superblock is a sequence parameter and has to be the same for 1 // sequence. In lightfield application, must choose the superblock size(either // 64x64 or 128x128) before the encoding starts. Otherwise, the default is // AOM_SUPERBLOCK_SIZE_DYNAMIC, and the superblock size will be set to 64x64 // internally. if (aom_codec_control(&codec, AV1E_SET_SUPERBLOCK_SIZE, AOM_SUPERBLOCK_SIZE_64X64)) die_codec(&codec, "Failed to set SB size"); u_blocks = (lf_width + lf_blocksize - 1) / lf_blocksize; v_blocks = (lf_height + lf_blocksize - 1) / lf_blocksize; reference_image_num = u_blocks * v_blocks; aom_img_fmt_t ref_fmt = AOM_IMG_FMT_I420; if (!CONFIG_LOWBITDEPTH) ref_fmt |= AOM_IMG_FMT_HIGHBITDEPTH; // Allocate memory with the border so that it can be used as a reference. for (i = 0; i < reference_image_num; i++) { if (!aom_img_alloc_with_border(&reference_images[i], ref_fmt, cfg->g_w, cfg->g_h, 32, 8, AOM_BORDER_IN_PIXELS)) { die("Failed to allocate image."); } } printf("\n Second pass: "******"Encoding Reference Images\n"); for (bv = 0; bv < v_blocks; ++bv) { for (bu = 0; bu < u_blocks; ++bu) { const int block_u_min = bu * lf_blocksize; const int block_v_min = bv * lf_blocksize; int block_u_end = (bu + 1) * lf_blocksize; int block_v_end = (bv + 1) * lf_blocksize; int u_block_size, v_block_size; int block_ref_u, block_ref_v; block_u_end = block_u_end < lf_width ? block_u_end : lf_width; block_v_end = block_v_end < lf_height ? block_v_end : lf_height; u_block_size = block_u_end - block_u_min; v_block_size = block_v_end - block_v_min; block_ref_u = block_u_min + u_block_size / 2; block_ref_v = block_v_min + v_block_size / 2; printf("A%d, ", (block_ref_u + block_ref_v * lf_width)); fseek(infile, (block_ref_u + block_ref_v * lf_width) * image_size_bytes, SEEK_SET); aom_img_read(raw, infile); get_raw_image(&frame_to_encode, raw, raw_shift); // Reference frames may be encoded without tiles. ++frame_count; printf("Encoding reference image %d of %d\n", bv * u_blocks + bu, u_blocks * v_blocks); encode_frame(&codec, frame_to_encode, frame_count, 1, AOM_EFLAG_NO_REF_LAST2 | AOM_EFLAG_NO_REF_LAST3 | AOM_EFLAG_NO_REF_GF | AOM_EFLAG_NO_REF_ARF | AOM_EFLAG_NO_REF_BWD | AOM_EFLAG_NO_REF_ARF2 | AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF | AOM_EFLAG_NO_UPD_ARF | AOM_EFLAG_NO_UPD_ENTROPY, writer); if (aom_codec_control(&codec, AV1_COPY_NEW_FRAME_IMAGE, &reference_images[frame_count - 1])) die_codec(&codec, "Failed to copy decoder reference frame"); } } cfg->large_scale_tile = 1; // Fixed q encoding for camera frames. cfg->rc_end_usage = AOM_Q; if (aom_codec_enc_config_set(&codec, cfg)) die_codec(&codec, "Failed to configure encoder"); // The fixed q value used in encoding. if (aom_codec_control(&codec, AOME_SET_CQ_LEVEL, 36)) die_codec(&codec, "Failed to set cq level"); if (aom_codec_control(&codec, AV1E_SET_FRAME_PARALLEL_DECODING, 1)) die_codec(&codec, "Failed to set frame parallel decoding"); if (aom_codec_control(&codec, AV1E_SET_SINGLE_TILE_DECODING, 1)) die_codec(&codec, "Failed to turn on single tile decoding"); // Set tile_columns and tile_rows to MAX values, which guarantees the tile // size of 64 x 64 pixels(i.e. 1 SB) for <= 4k resolution. if (aom_codec_control(&codec, AV1E_SET_TILE_COLUMNS, 6)) die_codec(&codec, "Failed to set tile width"); if (aom_codec_control(&codec, AV1E_SET_TILE_ROWS, 6)) die_codec(&codec, "Failed to set tile height"); for (bv = 0; bv < v_blocks; ++bv) { for (bu = 0; bu < u_blocks; ++bu) { const int block_u_min = bu * lf_blocksize; const int block_v_min = bv * lf_blocksize; int block_u_end = (bu + 1) * lf_blocksize; int block_v_end = (bv + 1) * lf_blocksize; int u, v; block_u_end = block_u_end < lf_width ? block_u_end : lf_width; block_v_end = block_v_end < lf_height ? block_v_end : lf_height; for (v = block_v_min; v < block_v_end; ++v) { for (u = block_u_min; u < block_u_end; ++u) { av1_ref_frame_t ref; ref.idx = 0; ref.use_external_ref = 1; ref.img = reference_images[bv * u_blocks + bu]; if (aom_codec_control(&codec, AV1_SET_REFERENCE, &ref)) die_codec(&codec, "Failed to set reference frame"); printf("C%d, ", (u + v * lf_width)); fseek(infile, (u + v * lf_width) * image_size_bytes, SEEK_SET); aom_img_read(raw, infile); get_raw_image(&frame_to_encode, raw, raw_shift); ++frame_count; printf("Encoding image %d of %d\n", frame_count - (u_blocks * v_blocks), lf_width * lf_height); encode_frame(&codec, frame_to_encode, frame_count, 1, AOM_EFLAG_NO_REF_LAST2 | AOM_EFLAG_NO_REF_LAST3 | AOM_EFLAG_NO_REF_GF | AOM_EFLAG_NO_REF_ARF | AOM_EFLAG_NO_REF_BWD | AOM_EFLAG_NO_REF_ARF2 | AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF | AOM_EFLAG_NO_UPD_ARF | AOM_EFLAG_NO_UPD_ENTROPY, writer); } } } } // Flush encoder. // No ARF, this should not be needed. while (encode_frame(&codec, NULL, -1, 1, 0, writer)) { } for (i = 0; i < reference_image_num; i++) aom_img_free(&reference_images[i]); if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec."); aom_video_writer_close(writer); printf("\nSecond pass complete. Processed %d frames.\n", frame_count); }
int main(int argc, char **argv) { aom_codec_ctx_t codec; AvxVideoReader *reader = NULL; AvxVideoWriter *writer = NULL; const AvxInterface *decoder = NULL; const AvxVideoInfo *info = NULL; int num_references; int i; aom_codec_pts_t pts; const char *tile_list_file = NULL; exec_name = argv[0]; if (argc != 5) die("Invalid number of arguments."); reader = aom_video_reader_open(argv[1]); if (!reader) die("Failed to open %s for reading.", argv[1]); num_references = (int)strtol(argv[3], NULL, 0); info = aom_video_reader_get_info(reader); // The writer to write out ivf file in tile list OBU, which can be decoded by // AV1 decoder. writer = aom_video_writer_open(argv[2], kContainerIVF, info); if (!writer) die("Failed to open %s for writing", argv[2]); tile_list_file = argv[4]; decoder = get_aom_decoder_by_fourcc(info->codec_fourcc); if (!decoder) die("Unknown input codec."); printf("Using %s\n", aom_codec_iface_name(decoder->codec_interface())); if (aom_codec_dec_init(&codec, decoder->codec_interface(), NULL, 0)) die_codec(&codec, "Failed to initialize decoder."); // Decode anchor frames. aom_codec_control_(&codec, AV1_SET_TILE_MODE, 0); printf("Reading %d reference images.\n", num_references); for (i = 0; i < num_references; ++i) { aom_video_reader_read_frame(reader); size_t frame_size = 0; const unsigned char *frame = aom_video_reader_get_frame(reader, &frame_size); pts = (aom_codec_pts_t)aom_video_reader_get_frame_pts(reader); // Copy references bitstream directly. if (!aom_video_writer_write_frame(writer, frame, frame_size, pts)) die_codec(&codec, "Failed to copy compressed anchor frame."); if (aom_codec_decode(&codec, frame, frame_size, NULL)) die_codec(&codec, "Failed to decode frame."); } // Decode camera frames. aom_codec_control_(&codec, AV1_SET_TILE_MODE, 1); aom_codec_control_(&codec, AV1D_EXT_TILE_DEBUG, 1); FILE *infile = aom_video_reader_get_file(reader); // Record the offset of the first camera image. const FileOffset camera_frame_pos = ftello(infile); printf("Loading compressed frames into memory.\n"); // Count the frames in the lightfield. int num_frames = 0; while (aom_video_reader_read_frame(reader)) { ++num_frames; } if (num_frames < 1) die("Input light field has no frames."); // Read all of the lightfield frames into memory. unsigned char **frames = (unsigned char **)malloc(num_frames * sizeof(unsigned char *)); size_t *frame_sizes = (size_t *)malloc(num_frames * sizeof(size_t)); // Seek to the first camera image. fseeko(infile, camera_frame_pos, SEEK_SET); for (int f = 0; f < num_frames; ++f) { aom_video_reader_read_frame(reader); size_t frame_size = 0; const unsigned char *frame = aom_video_reader_get_frame(reader, &frame_size); frames[f] = (unsigned char *)malloc(frame_size * sizeof(unsigned char)); memcpy(frames[f], frame, frame_size); frame_sizes[f] = frame_size; } printf("Read %d frames.\n", num_frames); // Copy first camera frame for getting camera frame header. This is done // only once. { size_t frame_size = frame_sizes[0]; const unsigned char *frame = frames[0]; pts = num_references; aom_tile_data frame_header_info = { 0, NULL, 0 }; // Need to decode frame header to get camera frame header info. So, here // decoding 1 tile is enough. aom_codec_control_(&codec, AV1_SET_DECODE_TILE_ROW, 0); aom_codec_control_(&codec, AV1_SET_DECODE_TILE_COL, 0); aom_codec_err_t aom_status = aom_codec_decode(&codec, frame, frame_size, NULL); if (aom_status) die_codec(&codec, "Failed to decode tile."); aom_codec_control_(&codec, AV1D_GET_FRAME_HEADER_INFO, &frame_header_info); size_t obu_size_offset = (uint8_t *)frame_header_info.coded_tile_data - frame; size_t length_field_size = frame_header_info.coded_tile_data_size; // Remove ext-tile tile info. uint32_t frame_header_size = (uint32_t)frame_header_info.extra_size - 1; size_t bytes_to_copy = obu_size_offset + length_field_size + frame_header_size; unsigned char *frame_hdr_buf = (unsigned char *)malloc(bytes_to_copy); if (frame_hdr_buf == NULL) die_codec(&codec, "Failed to allocate frame header buffer."); memcpy(frame_hdr_buf, frame, bytes_to_copy); // Update frame header OBU size. size_t bytes_written = 0; if (aom_uleb_encode_fixed_size( frame_header_size, length_field_size, length_field_size, frame_hdr_buf + obu_size_offset, &bytes_written)) die_codec(&codec, "Failed to encode the tile list obu size."); // Copy camera frame header bitstream. if (!aom_video_writer_write_frame(writer, frame_hdr_buf, bytes_to_copy, pts)) die_codec(&codec, "Failed to copy compressed camera frame header."); free(frame_hdr_buf); } // Read out the image format. aom_img_fmt_t ref_fmt = 0; if (aom_codec_control(&codec, AV1D_GET_IMG_FORMAT, &ref_fmt)) die_codec(&codec, "Failed to get the image format"); const int bps = get_image_bps(ref_fmt); if (!bps) die_codec(&codec, "Invalid image format."); // read out the tile size. unsigned int tile_size = 0; if (aom_codec_control(&codec, AV1D_GET_TILE_SIZE, &tile_size)) die_codec(&codec, "Failed to get the tile size"); const unsigned int tile_width = tile_size >> 16; const unsigned int tile_height = tile_size & 65535; // Allocate a buffer to store tile list bitstream. const size_t data_sz = MAX_TILES * ALIGN_POWER_OF_TWO(tile_width, 5) * ALIGN_POWER_OF_TWO(tile_height, 5) * bps / 8; unsigned char *tl_buf = (unsigned char *)malloc(data_sz); if (tl_buf == NULL) die_codec(&codec, "Failed to allocate tile list buffer."); aom_codec_pts_t tl_pts = num_references; const uint8_t output_frame_width_in_tiles_minus_1 = output_frame_width / tile_width - 1; const uint8_t output_frame_height_in_tiles_minus_1 = output_frame_height / tile_height - 1; printf("Reading tile list from file.\n"); char line[1024]; FILE *tile_list_fptr = fopen(tile_list_file, "r"); if (!tile_list_fptr) die_codec(&codec, "Failed to open tile list file."); int num_tiles = 0; TILE_LIST_INFO tiles[MAX_TILES]; while ((fgets(line, 1024, tile_list_fptr)) != NULL) { if (line[0] == 'F' || num_tiles >= MAX_TILES) { // Flush existing tile list and start another, either because we hit a // new render frame or because we've hit our max number of tiles per list. if (num_tiles > 0) { process_tile_list(tiles, num_tiles, tl_pts, frames, frame_sizes, &codec, tl_buf, writer, output_frame_width_in_tiles_minus_1, output_frame_height_in_tiles_minus_1); ++tl_pts; } num_tiles = 0; } if (line[0] == 'F') { continue; } if (sscanf(line, "%d %d %d %d", &tiles[num_tiles].image_idx, &tiles[num_tiles].reference_idx, &tiles[num_tiles].tile_col, &tiles[num_tiles].tile_row) == 4) { if (tiles[num_tiles].image_idx >= num_frames) { die("Tile list image_idx out of bounds: %d >= %d.", tiles[num_tiles].image_idx, num_frames); } if (tiles[num_tiles].reference_idx >= num_references) { die("Tile list reference_idx out of bounds: %d >= %d.", tiles[num_tiles].reference_idx, num_references); } ++num_tiles; } } if (num_tiles > 0) { // Flush out the last tile list. process_tile_list(tiles, num_tiles, tl_pts, frames, frame_sizes, &codec, tl_buf, writer, output_frame_width_in_tiles_minus_1, output_frame_height_in_tiles_minus_1); ++tl_pts; } const int num_tile_lists = (int)(tl_pts - pts); printf("Finished processing tile lists. Num tile lists: %d.\n", num_tile_lists); free(tl_buf); for (int f = 0; f < num_frames; ++f) { free(frames[f]); } free(frame_sizes); free(frames); if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec"); aom_video_writer_close(writer); aom_video_reader_close(reader); return EXIT_SUCCESS; }