Skip to content

Instantly share code, notes, and snippets.

@Birch-san
Created July 27, 2023 23:07
Show Gist options
  • Star 2 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save Birch-san/1a4db45aef05d1a8c9cade049d939399 to your computer and use it in GitHub Desktop.
Save Birch-san/1a4db45aef05d1a8c9cade049d939399 to your computer and use it in GitHub Desktop.
Computing aspect ratio buckets
import numpy as np
import math
from numpy.typing import NDArray
# we are trying to make buckets of varying aspect ratios,
# all with about the same area (equivalent to a 512x512 square)
square_side = 512
buckets = 8
widest_aspect: float = math.atan2(1, 2) # 1/2 = 0.5 aspect ratio
tallest_aspect: float = math.atan2(2, 1) # 2/1 = 2.0 aspect ratio
thetas: NDArray = np.linspace(widest_aspect, tallest_aspect, buckets)
hypotenuses: NDArray = square_side/(2*np.sin(thetas)*np.cos(thetas))**.5
heights: NDArray = hypotenuses*2**.5*np.sin(thetas)
widths: NDArray = hypotenuses*2**.5*np.cos(thetas)
buckets: NDArray = np.column_stack([heights, widths])
# all buckets will have about the same area as square_side**2
# you can verify this by multiplying their widths by their heights:
# np.multiply.reduce(buckets, axis=-1)
# round to nearest 8px (e.g. to be suitable for encoding in VAE)
buckets_rounded: NDArray = 8*np.rint(buckets / 8).astype(int)
pass
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment