How to read video frames in hadoop?如何在Hadoop中讀取視頻幀？

To process specialized file formats (such as video) in Hadoop, you'd have to write a custom InputFormat and RecordReader that understands how to turn a video file into splits (the InputFormat) and then read splits into values (the RecordReader).在Hadoop要處理的專用文件格式（如視頻），你就必須寫一個自定義的InputFormat和RecordReader，瞭解如何將一個視頻文件分割（InputFormat），而後讀值（RecordReader）分裂成。 This is a non-trivial task and requires some intermediate knowledge of how Hadoop handles the splitting of data.這是一個不平凡的任務和Hadoop的處理分割的數據須要一些中間的知識。 I highly recommend Tom White's Hadoop the Definitive Guide book by O'Reilly as well as the videos on http://www.cloudera.com .我強烈建議湯姆白色的Hadoop權威指南書由O'Reilly和視頻http://www.cloudera.com 。 (Full disclosure: I work for Cloudera.) （披露：我工做的Cloudera的。）

Keep in mind that video formats are generally compressed which gets even more complicated because InputSplits (created by an InputFormat) are simple byte offsets into the file (normally).請記住，通常都是壓縮的視頻格式變得更加複雜，由於InputSplits建立一個InputFormat是簡單的字節偏移量到文件中（一般狀況下）。 Start withhttp://hadoop.apache.org/common/docs/current/api/org/apache/hadoop/mapred/InputFormat.html從與http://hadoop.apache.org/common/docs/current/api/org/apache/hadoop/mapred/InputFormat.html開始

To summarize: InputFormat knows how to generate a list of InputSplit objects that are (usually) between 64MB and 128MB and do NOT respect the notion of frames.總結：InputFormat知道如何生成的列表InputSplit對象（一般狀況下），64MB和128MB之間， 不尊重的概念框架。 The RecordReader then is used to read frames out of a InputSplit to create value objects that the map reduce job can process. 「的RecordReader而後是用於讀取幀一的InputSplit，以建立的Map Reduce做業能夠處理的值對象。 If you want to generate video output you'll also need to write a custom OutputFormat.若是你想生成視頻輸出，您還須要編寫一個自定義的OutputFormat。

Hope this helps.但願這會有所幫助。