libaom 源码分析：scalable_decoder.c 文件

DogDaoDao2025-03-19 14:22

libaom

  1. 基本特性

    • 开放和免版税：libaom 提供了一个开放源代码的编码器，任何个人和组织都可以免费使用，无需支付版税，这促进了它在各种应用中的广泛采用。
    • 高效的编码：旨在提供高效的视频压缩，以适应不同的网络条件和设备性能。AV1 编解码器通常比 HEVC（H.265）提供约 30% 更好的压缩效率，在相同质量下可以减少带宽消耗。
    • 先进的压缩技术：实现了 AV1 编解码器，使用了一系列先进的压缩技术，如 CDEF（Constrained Directional Enhancement Filtering）、CIC（Compound Internal Coding）和 PAET（Probabilistic Angular Early Termination）等。
    • 可配置性：提供了多种配置选项，允许开发者根据应用需求调整编码参数。
    • 实时编码支持：支持实时编码，适用于直播和实时通信应用。
    • 社区支持：作为一个开源项目，得到了活跃的社区支持，不断有新功能和改进被加入。

  2. 应用场景

    libaom 可以用于各种需要视频编解码的场景，如在线视频播放、视频会议、视频存储、流媒体、网页视频等。许多视频播放器、浏览器和视频服务提供商都采用了 libaom 来实现 AV1 视频编码标准。

  3. 性能提升

    • 压缩性能：经过团队的研究和尝试，AV1 的压缩性能在过去几年中有了显著的提高，相对于 VP9 而言，在以 PSNR 或 SSIM 这样的客观指标条件下能提高 36% 以上。
    • 编码器速度和内存需求：相对于 2018 年 bit - stream finalization 的时候，Libaom 编码器的速度提升 150 倍以上，在 4K 的视频压缩条件下对内存的需求下降了 80%。同时，Libaom 支持不同的 speed setting，用户可以根据实际需求进行选择，在速度和压缩性能之间进行平衡。

scalable_decoder.c 介绍

  1. 功能描述 ：该文件是一个可伸缩解码器的示例，输入文件包含两层空间层的压缩数据，格式为 OBU，数据通过解码器处理，解码后的帧会写入磁盘，基本层和增强层分别存储为 out_lyr0.yuv 和 out_lyr1.yuv 文件。
  2. 文件位置：libaom/examples/scalable_decoder.c
  3. main 函数流程梳理：
bash 复制代码 
. 初始化阶段
   - 解析命令行参数，检查输入文件
   - 打开输入文件并初始化 OBU 解码上下文
   - 获取 AV1 解码器接口
   - 初始化解码器上下文
   - 设置解码器输出所有层的控制参数

. 文件验证
   - 读取文件头信息，验证是否为有效的 OBU 文件
   - 获取视频流的空间层数量

. 输出文件准备
   - 打开基本层输出文件 `out_lyr0.yuv`
   - 根据空间层数量打开对应的增强层输出文件

. 解码循环
   - 读取时间单元数据
   - 解码每一帧
   - 获取解码后的图像
   - 对图像进行位深转换
   - 根据空间层 ID 将图像写入对应的输出文件
   - 更新帧计数器

. 清理阶段
   - 销毁解码器上下文
   - 关闭所有输出文件
   - 关闭输入文件
  1. 文件内部调用的 API 解析：
  • 实例化解码器：get_aom_decoder_by_index(0) 通过索引 0 获取 AV1 解码器接口，aom_codec_iface_name(decoder) 获取解码器的名称；
c 复制代码 
  aom_codec_iface_t *decoder = get_aom_decoder_by_index(0);
  printf("Using %s\n", aom_codec_iface_name(decoder));
  • 初始化：aom_codec_ctx_t codec声明一个解码器上下文变量 codec ，用于存储解码器的状态信息；aom_codec_dec_init() 初始化解码器；
c 复制代码 
  aom_codec_ctx_t codec;
  if (aom_codec_dec_init(&codec, decoder, NULL, 0))
    die("Failed to initialize decoder.");
  • 编码器能力控制 ：aom_codec_control() 用于设置解码器的控制参数；
    AV1D_SET_OUTPUT_ALL_LAYERS 控制参数，表示是否输出所有层；
c 复制代码 
 if (aom_codec_control(&codec, AV1D_SET_OUTPUT_ALL_LAYERS, 1)) {
    die_codec(&codec, "Failed to set output_all_layers control.");
  }
  • 输入流的OBU头信息解析：aom_codec_peek_stream_info解析空间层配置信息；
c 复制代码 
  // peak sequence header OBU to get number of spatial layers
  const size_t ret = fread(tmpbuf, 1, 32, inputfile);
  if (ret != 32) die_codec(&codec, "Input is not a valid obu file");
  si.is_annexb = 0;
  if (aom_codec_peek_stream_info(decoder, tmpbuf, 32, &si)) {
    die_codec(&codec, "Input is not a valid obu file");
  }
  fseek(inputfile, -32, SEEK_CUR);
  • 文件校验和输出文件准备：
c 复制代码 
 if (!file_is_obu(&obu_ctx))
    die_codec(&codec, "Input is not a valid obu file");

  // open base layer output yuv file
  snprintf(filename, sizeof(filename), "out_lyr%d.yuv", 0);
  if (!(outfile[0] = fopen(filename, "wb")))
    die("Failed top open output for writing.");

  // open any enhancement layer output yuv files
  for (i = 1; i < si.number_spatial_layers; i++) {
    snprintf(filename, sizeof(filename), "out_lyr%u.yuv", i);
    if (!(outfile[i] = fopen(filename, "wb")))
      die("Failed to open output for writing.");
  }
  • 解码核心逻辑：obudec_read_temporal_unit一直读取OBU单元，aom_codec_decode解码的核心函数，aom_codec_get_frame获取解码帧，aom_img_downshift将原图像进行下采样，aom_img_write根据不同的spatial_id写入输出视频文件；
c 复制代码 
  while (!obudec_read_temporal_unit(&obu_ctx, &buf, &bytes_in_buffer,
                                    &buffer_size)) {
    aom_codec_iter_t iter = NULL;
    aom_image_t *img = NULL;
    if (aom_codec_decode(&codec, buf, bytes_in_buffer, NULL))
      die_codec(&codec, "Failed to decode frame.");

    while ((img = aom_codec_get_frame(&codec, &iter)) != NULL) {
      aom_image_t *img_shifted =
          aom_img_alloc(NULL, AOM_IMG_FMT_I420, img->d_w, img->d_h, 16);
      img_shifted->bit_depth = 8;
      aom_img_downshift(img_shifted, img,
                        img->bit_depth - img_shifted->bit_depth);
      if (img->spatial_id == 0) {
        printf("Writing        base layer 0 %d\n", frame_cnt);
        aom_img_write(img_shifted, outfile[0]);
      } else if (img->spatial_id <= (int)(si.number_spatial_layers - 1)) {
        printf("Writing enhancement layer %d %d\n", img->spatial_id, frame_cnt);
        aom_img_write(img_shifted, outfile[img->spatial_id]);
      } else {
        die_codec(&codec, "Invalid bitstream. Layer id exceeds layer count");
      }
      if (img->spatial_id == (int)(si.number_spatial_layers - 1)) ++frame_cnt;
    }
  }
  • 资源销毁和回收清理工作：
c 复制代码 
  printf("Processed %d frames.\n", frame_cnt);
  if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec");

  for (i = 0; i < si.number_spatial_layers; i++) fclose(outfile[i]);

  fclose(inputfile);
  1. 源码：
c 复制代码 
/*
 * Copyright (c) 2018, Alliance for Open Media. All rights reserved.
 *
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 */

// Scalable Decoder
// ==============
//
// This is an example of a scalable decoder loop. It takes a 2-spatial-layer
// input file
// containing the compressed data (in OBU format), passes it through the
// decoder, and writes the decompressed frames to disk. The base layer and
// enhancement layers are stored as separate files, out_lyr0.yuv and
// out_lyr1.yuv, respectively.
//
// Standard Includes
// -----------------
// For decoders, you only have to include `aom_decoder.h` and then any
// header files for the specific codecs you use. In this case, we're using
// av1.
//
// Initializing The Codec
// ----------------------
// The libaom decoder is initialized by the call to aom_codec_dec_init().
// Determining the codec interface to use is handled by AvxVideoReader and the
// functions prefixed with aom_video_reader_. Discussion of those functions is
// beyond the scope of this example, but the main gist is to open the input file
// and parse just enough of it to determine if it's a AVx file and which AVx
// codec is contained within the file.
// Note the NULL pointer passed to aom_codec_dec_init(). We do that in this
// example because we want the algorithm to determine the stream configuration
// (width/height) and allocate memory automatically.
//
// Decoding A Frame
// ----------------
// Once the frame has been read into memory, it is decoded using the
// `aom_codec_decode` function. The call takes a pointer to the data
// (`frame`) and the length of the data (`frame_size`). No application data
// is associated with the frame in this example, so the `user_priv`
// parameter is NULL. The `deadline` parameter is left at zero for this
// example. This parameter is generally only used when doing adaptive post
// processing.
//
// Codecs may produce a variable number of output frames for every call to
// `aom_codec_decode`. These frames are retrieved by the
// `aom_codec_get_frame` iterator function. The iterator variable `iter` is
// initialized to NULL each time `aom_codec_decode` is called.
// `aom_codec_get_frame` is called in a loop, returning a pointer to a
// decoded image or NULL to indicate the end of list.
//
// Processing The Decoded Data
// ---------------------------
// In this example, we simply write the encoded data to disk. It is
// important to honor the image's `stride` values.
//
// Cleanup
// -------
// The `aom_codec_destroy` call frees any memory allocated by the codec.
//
// Error Handling
// --------------
// This example does not special case any error return codes. If there was
// an error, a descriptive message is printed and the program exits. With
// few exceptions, aom_codec functions return an enumerated error status,
// with the value `0` indicating success.

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "aom/aom_decoder.h"
#include "aom/aomdx.h"
#include "common/obudec.h"
#include "common/tools_common.h"
#include "common/video_reader.h"

static const char *exec_name;

#define MAX_LAYERS 5

void usage_exit(void) {
  fprintf(stderr, "Usage: %s <infile>\n", exec_name);
  exit(EXIT_FAILURE);
}

int main(int argc, char **argv) {
  int frame_cnt = 0;
  FILE *outfile[MAX_LAYERS];
  char filename[80];
  FILE *inputfile = NULL;
  uint8_t *buf = NULL;
  size_t bytes_in_buffer = 0;
  size_t buffer_size = 0;
  struct AvxInputContext aom_input_ctx;
  struct ObuDecInputContext obu_ctx = { &aom_input_ctx, NULL, 0, 0, 0 };
  aom_codec_stream_info_t si;
  uint8_t tmpbuf[32];
  unsigned int i;

  exec_name = argv[0];

  if (argc != 2) die("Invalid number of arguments.");

  if (!(inputfile = fopen(argv[1], "rb")))
    die("Failed to open %s for read.", argv[1]);
  obu_ctx.avx_ctx->file = inputfile;
  obu_ctx.avx_ctx->filename = argv[1];

  aom_codec_iface_t *decoder = get_aom_decoder_by_index(0);
  printf("Using %s\n", aom_codec_iface_name(decoder));

  aom_codec_ctx_t codec;
  if (aom_codec_dec_init(&codec, decoder, NULL, 0))
    die("Failed to initialize decoder.");

  if (aom_codec_control(&codec, AV1D_SET_OUTPUT_ALL_LAYERS, 1)) {
    die_codec(&codec, "Failed to set output_all_layers control.");
  }

  // peak sequence header OBU to get number of spatial layers
  const size_t ret = fread(tmpbuf, 1, 32, inputfile);
  if (ret != 32) die_codec(&codec, "Input is not a valid obu file");
  si.is_annexb = 0;
  if (aom_codec_peek_stream_info(decoder, tmpbuf, 32, &si)) {
    die_codec(&codec, "Input is not a valid obu file");
  }
  fseek(inputfile, -32, SEEK_CUR);

  if (!file_is_obu(&obu_ctx))
    die_codec(&codec, "Input is not a valid obu file");

  // open base layer output yuv file
  snprintf(filename, sizeof(filename), "out_lyr%d.yuv", 0);
  if (!(outfile[0] = fopen(filename, "wb")))
    die("Failed top open output for writing.");

  // open any enhancement layer output yuv files
  for (i = 1; i < si.number_spatial_layers; i++) {
    snprintf(filename, sizeof(filename), "out_lyr%u.yuv", i);
    if (!(outfile[i] = fopen(filename, "wb")))
      die("Failed to open output for writing.");
  }

  while (!obudec_read_temporal_unit(&obu_ctx, &buf, &bytes_in_buffer,
                                    &buffer_size)) {
    aom_codec_iter_t iter = NULL;
    aom_image_t *img = NULL;
    if (aom_codec_decode(&codec, buf, bytes_in_buffer, NULL))
      die_codec(&codec, "Failed to decode frame.");

    while ((img = aom_codec_get_frame(&codec, &iter)) != NULL) {
      aom_image_t *img_shifted =
          aom_img_alloc(NULL, AOM_IMG_FMT_I420, img->d_w, img->d_h, 16);
      img_shifted->bit_depth = 8;
      aom_img_downshift(img_shifted, img,
                        img->bit_depth - img_shifted->bit_depth);
      if (img->spatial_id == 0) {
        printf("Writing        base layer 0 %d\n", frame_cnt);
        aom_img_write(img_shifted, outfile[0]);
      } else if (img->spatial_id <= (int)(si.number_spatial_layers - 1)) {
        printf("Writing enhancement layer %d %d\n", img->spatial_id, frame_cnt);
        aom_img_write(img_shifted, outfile[img->spatial_id]);
      } else {
        die_codec(&codec, "Invalid bitstream. Layer id exceeds layer count");
      }
      if (img->spatial_id == (int)(si.number_spatial_layers - 1)) ++frame_cnt;
    }
  }

  printf("Processed %d frames.\n", frame_cnt);
  if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec");

  for (i = 0; i < si.number_spatial_layers; i++) fclose(outfile[i]);

  fclose(inputfile);

  return EXIT_SUCCESS;
}
相关推荐
AI技术控3 小时前
扩散模型算法实战——视频生成（主页有源码）
算法·音视频
背水4 小时前
音频大语言模型可作为描述性语音质量评价器
人工智能·语言模型·音视频·语音识别
Aomnitrix4 小时前
Qt 实操记录:打造自己的“ QQ 音乐播放器”
开发语言·c++·qt·ui·音视频
智联视频超融合平台4 小时前
大华HTTP协议在智联视频超融合平台中的接入方法
网络协议·http·音视频·实时音视频·视频编解码
爱做ppt的阿伟4 小时前
一键爬取b站视频
python·音视频
Bill Adams9 小时前
开源免费一句话生成儿童故事视频核心思想解析
人工智能·开源·音视频·视频生成
小gpt&12 小时前
03 介绍ffmpeg 视频解码流程
c++·qt·ffmpeg·音视频
找方案1 天前
短视频下载去水印，用什么工具好？
音视频·去水印工具
热门推荐
01命令行方式将mysql数据库迁移到达梦数据库（全步骤）02从零安装 LLaMA-Factory 微调 Qwen 大模型成功及所有的坑03DeepSeek各版本说明与优缺点分析04本地部署DeepSeek教程（Mac版本）05如何在WPS和Word/Excel中直接使用DeepSeek功能06本地化部署AI知识库：基于Ollama+DeepSeek+AnythingLLM保姆级教程07四大AI编程工具组合测评08DeepSeek RAGFlow构建本地知识库系统09DeepSeek R1本地化部署 Ollama + Chatbox 打造最强 AI 工具10计算机视觉 CV 八股分享 [自用]（更新中......）