Mirror of forum.nim-lang.org

6129 :: Arraymancer - kernel

• mantielero (orginal) [2020-03-28T21:11:40+01:00] view original

  I was wondering if it is possible to apply a small kernel to a bigger matrix. I have implemented a small blur filter, and I would like to see how it feels to do it with arraymancer.

---

• mratsim (orginal) [2020-04-03T16:21:36+02:00] view original

  It's not compatible anymore but that is how you can do it: https://github.com/edubart/arraymancer-vision/blob/master/src/arraymancer_vision/filters.nim

  That said, for convolution / cross-correlation filters like blur you can use conv2D: https://github.com/mratsim/Arraymancer/blob/master/src/nn_primitives/nnp_convolution.nim#L28-L31

  • you can use a zero bias for image processing (it's values that are added to the result of the convolution).
  • it uses a batch of images as input not a single one (format NCHW, N image number, C color channels, H height, W width) there is a converter to convert from HWC to CHW
  • Reading images is done here: https://github.com/mratsim/Arraymancer/blob/0605c7fcd34e216623a891f4963d2a4ef98882b1/src/io/io_image.nim#L20-L54
  • you can myImage.read_image().unsqueeze(0) to have the image in the proper format for convolution.

---