Harshit Sikchi
Harshit Sikchi
I 'll look into it and make the changes
@domfarolino I have changed the commit message as well as the indentation.Can you check and see if its okay?
Maybe this will help: https://github.com/CharlesShang/FastMaskRCNN/issues/26
I experienced this kind of difficulties in all sparse reward setting. Is ARS a good way to go for these optimization landscapes?
Having a trait function would be definitely better. I'll make those changes and update the pull request. Yes, for now it is specific to 2D. I'll see if I can...
I have updated the commit with support for N-Dimensional inputs also added the uses_histogram function to maintain the generic implementation for mapwindow. I had some doubts about the code:- 1....
I agree, I am converting the code to column major for now. Let's see whether reinitializing the histogram can make it faster.
@tejus-gupta I have made changes in the code to be column major and obtained significant time reduction. Also, zig-zag traversal increases the time complexity. In the paper, for constant time...
I am on a flight, will fix this as soon as I reach back.
I have changed the mapwindow dispatch function to work for types img::Union{AbstractArray{N0f8,N},AbstractArray{ColorTypes.Gray{N0f8},N}} and removed type incompatibilities. Still, when I am trying to make it run in O(n) complexity by breaking...