【HM12.0】Data structure and access method

翻译 2013年12月04日 10:50:43
• LCU vs. Sub-CU
All CUs are handled throughTComDataCU* pointer(所有CU是通过 TComDataCU *指针操作处理)
From the LCU (largest CU), Sub-CUis processed recursively by z-scan(CU通过Z顺序扫描)



• TComDataCU class has all information we need 
  – Actual data storage is allocated only at LCU (largest CU) level
    • In Sub-CUs, data storage points to suitable position in LCU storage


• Every information in CU is accessed by index (CU中信息的访问方式依靠索引号
   –Basically, every index is the storage unit number within the CU
   –In source-code,AbsPartIdx means absolute index in the given CU (代码中AbsPartIdx代表CU内单元的索引号)


 How to obtain absolute index in LCU?(LCU内单元的索引获取方式(通过m_uiAbsIdxInLCU))
  – EachTComDataCU class hasm_uiAbsIdxInLCU, which means the absolute z-scanindex in LCU basis
  – Useful to derive index ofneighbouring CUs


 • How to convert between z-scan and raster scan in LCU-basedindex?(Z扫描与光栅扫描转换)
   –Two arrays,g_auiZscanToRaster andg_auiRasterToZscan
   –g_auiZscanToRaster[ z-scan index ] = raster scanindex
   –g_auiRasterToZscan[ raster index ] = z-scan index
    Raster scan is useful to derive index ofneighbouring CUs


• Transform coefficient data structure (变换后系数数据结构)
   –1D arrays,m_pcTrCoeffY,m_pcTrCoeffCb andm_pcTrCoeffCr are allocated in LCU
      For each CU, getCoeffY, getCoeffCb and getCoeffCr give the starting point of coefficients. They are arranged         using  raster-scan



相关文章推荐

[LeetCode]Add and Search Word - Data structure design

字典树( Trie ) 推荐一个对字典树解释的很好的博客 Trie树:应用于统计和排序  解题思路: 阅读完题目,大家都知道是字典树。对于树的搜索算法有 深度优先搜索算法(dfs)和...

algorithm and data structure in C

  • 2007-09-02 08:34
  • 4.49MB
  • 下载

LeetCode211. Add and Search Word - Data structure design前缀树(Trie,字典树)实战

Design a data structure that supports the following two operations: void addWord(word) bool search(...

Leetcode211 - Add and Search Word - Data structure design

我的新博客地址:HuHaoyu's Blog,最近新增了Jekyll的paginator分页功能和Swiftype的搜索引擎。欢迎参观!
内容举报
返回顶部
收藏助手
不良信息举报
您举报文章:深度学习:神经网络中的前向传播和反向传播算法推导
举报原因:
原因补充:

(最多只允许输入30个字)