i think that the example 9-4 "separable Convolution 3x3 by One Row" in IPP vol2 can also beimplemented with ippiFilterRow/Column. but i do not clearly know what the advantages of using ippiFilterRow/ColumnPipeline are
and what i should pay extra attention to when using ippiFilterRow/ColumnPipeline instead of ippiFilterRow/Column.
can someone give me explanation in detail?
are there some relevant documentations that could be recommended for me to understand this?
thanks in advance.
Sorry, Justfound the questionwas still on open.
Yes, both APIs are identical by functionality and can be substituted by each other. ippiFilterRow/ColumnPipeline belong to ippCV (computer vision) domain. Difference in APIs is that the pipeline API version designed to effective work in pipeline mode where whole image processed by parts through all operations in pipeline. In pipeline processing by image parts, every image part should be wrapped with border before filter invocation. For that purpose ippiFilterRow/ColumnPipeline functions work with array of pointers to the image rows/columns while ippiFilterRow/Column require whole 2D image buffer. Array of pointers in ippiFilterRow/ColumnPipeline allows more effectively create image border (just row pointers substitution instead of image copying) and this is resulted in memory and calculation saving.