I have a situation where I am augmenting some machine learning data and I want to do an affine tranformation followed by a warp (remap). Is there a way of doing this with a single function call? Single milliseconds are important to me, so I would prefer not to make two separate calls.
It turns out that all the lenses I use have essentially the exact same warping coefficients, so I can reuse the map. Furthermore I am doing a different affine transformation on every image. The point is, the strategy of building a map that bakes in the affine transformation is not optimal in this situation.
It seems likely that avoiding the extra memory buffer and Interpolation steps will be faster overall. However, I don’t know of any opencv endpoints for doing this. Does this already exist? If not, how to I propose it?