imresize_and_pad_if_need

imresize_and_pad_if_need(img: np.ndarray, max_h: int, max_w: int, interpolation: str | int | INTER = INTER.BILINEAR, pad_value: int | tuple[int, int, int] | None = 0, pad_mode: str | int | BORDER = BORDER.CONSTANT, return_scale: bool = False) -> np.ndarray | tuple[np.ndarray, float]

• Description: Resizes an image to fit within (max_h, max_w), and pads to the fixed target size when needed.

• Parameters

  • img (np.ndarray): Input image.
  • max_h (int): Max output height (and the fixed padded height).
  • max_w (int): Max output width (and the fixed padded width).
  • interpolation (str | int | INTER): Resize interpolation. Default is INTER.BILINEAR.
  • pad_value (int | tuple[int, int, int] | None): Padding value. For 3-channel images, you can pass an int or a tuple (OpenCV convention: BGR). Default is 0.
  • pad_mode (str | int | BORDER): Padding mode. Default is BORDER.CONSTANT.
  • return_scale (bool): Whether to return the resize scale. Default is False.

• Returns

  • When return_scale=False: returns np.ndarray.
  • When return_scale=True: returns (np.ndarray, float), where float is scale = min(max_h/raw_h, max_w/raw_w).

• Notes

  • Padding is applied to bottom and right only (top=0, left=0).
  • The image will be downscaled when max_h/max_w is smaller than the original, and upscaled when larger.

• Example

  import capybara as cb
  
  img = cb.imread('lena.png')
  
  out, scale = cb.imresize_and_pad_if_need(
      img,
      max_h=640,
      max_w=640,
      pad_value=0,
      return_scale=True,
  )
  print(out.shape, scale)