🌀 pepedpid

pepedpid is a Rust implementation of Rapid, Detail-Preserving Image Downscaling (DPID), designed to be used as a Python library via rust-py bindings. It combines high performance with ease of use in Python projects, such as machine learning pipelines or image processing tasks.

🚀 Quick Start

Installation:

pip install pepeline pepedpid

Usage example:

from pepeline import read, save, ImgFormat
from pepedpid import dpid_resize,cubic_resize

# Load an image in f32 format (normalized [0,1])
img = read("test.png", format=ImgFormat.F32)

# Apply DPID resizing
img_dpid = dpid_resize(img, 512, 512, 0.5)
img_matlab_bicubic = cubic_resize(img,512,512)
# Save the result
save(img_dpid, "img_dpid.png")
save(img_matlab_bicubic, "img_matlab_bicubic.png")

⚙️ Arguments for dpid_resize

dpid_resize(input: np.ndarray, h: int, w: int, l: float) -> np.ndarray

Parameters:

• input (np.ndarray) — input image of type float32, normalized in the range [0.0, 1.0]. Expected shape: (H, W, C), where C = 1 (grayscale) or 3 (RGB).

• h (int) — target height of the image.

• w (int) — target width of the image.

• l (float) — the λ coefficient, controlling the trade-off between smoothing and detail preservation:

  • λ ≈ 0.0 — maximum smoothing, the image will be soft.

  • λ ≈ 1.0 — maximum detail preservation, resulting in a sharp image.

  • Recommended value: 0.5 — balance between smoothness and sharpness.

Returns:

• np.ndarray — the downscaled image (float32, range [0.0, 1.0], shape (h, w, C)).