Did Fast.ai in 1.0 start to use PIL library instead of OpenCV (like it was in versions 0.x) for working with images? What were the reasons for that?
In lesson 1 of part 2 Jeremy explains why OpenCV is a better library than PIL or Torch Vision, but then it was replaced to PIL.
1 Like
This is discussed briefly here. There are also some benchmark liked showing that PIL is a lot faster with Pillow SIMD implementation. This SIMD implementation is the main reason for the speed.
1 Like