I’ve an image form pointcloud which is sparse, Using UNet can be an issue because there will not be many points in its window.
Its pixel values are basically classes. e.g. [3,4]=11 where pixel 3,4 has class 11.
My image looks like
I’ve tried using
imgg = imfilter(img, Kernel.gaussian(3))
but that will change classification assigned per pixel.
Is there an increase resolution function in julia? Thanks
If it is important that the output pixels have classes drawn from the input labels, perhaps you could use a nearest-neighbor interpolation. Assuming your pointcloud is stored as an array of
points[N, 3] where the first two columns are the coordinates and
points[1:end, 3] are the labels:
itp = interpolate(
(points[1:end, 1], points[1:end, 2]),
# itp can be queried by (x,y) to retrieve the class at a particular position,
# perhaps in a loop to generate an output image at a desired resolution.
Thanks, i’ll give it a try
Using a higher meters per pixel setting for the point cloud reduces the resolution and the gaps