Resize

The Resize module provides GPU-accelerated image scaling with multiple interpolation algorithms.

All resize methods return a new image at the target dimensions.

Aspect Ratio Preservation

Specify only width or height to maintain aspect ratio automatically.

Example:

from pyimagecuda import load, Resize, save

img = load("photo.jpg")  # 1920×1080

# Scale to width 800
resized = Resize.lanczos(img, width=800)
# Result: 800×450

save(resized, 'output.jpg')

Right-click to download and compare in full size

To force specific dimensions (may distort), specify both width and height.

Interpolation Methods

Nearest Neighbor

Fastest method with no interpolation (blocky results).

Example:

from pyimagecuda import load, Resize, save

img = load("photo.jpg")

scaled = Resize.nearest(img, width=128, height=128)

save(scaled, 'output.png')

Right-click to download and compare in full size

Use for: Pixel art, quick previews, integer scaling

Quality: Low | Speed: Fastest

Bilinear

Fast linear interpolation (smooth results).

Example:

from pyimagecuda import load, Resize, save

img = load("photo.jpg")

# Fast general-purpose resize
resized = Resize.bilinear(img, width=800)

save(resized, 'output.jpg')

Right-click to download and compare in full size

Use for: General purpose scaling, real-time applications

Quality: Medium | Speed: Fast

Bicubic

High-quality cubic interpolation.

Example:

from pyimagecuda import load, Resize, save

img = load("photo.jpg")

# High-quality upscale
resized = Resize.bicubic(img, width=1920)

save(resized, 'output.jpg')

Right-click to download and compare in full size

Use for: High-quality upscaling, photography, print materials

Quality: High | Speed: Medium

Lanczos

Highest quality with sharp details (recommended).

Example:

from pyimagecuda import load, Resize, save

img = load("photo.jpg")

# Maximum quality upscale
resized = Resize.lanczos(img, width=3840)

save(resized, 'output.jpg')

Right-click to download and compare in full size

Use for: Maximum quality upscaling, professional photography

Quality: Highest | Speed: Slower (still fast on GPU)

Parameters

All methods share the same parameters:

src (Image): Source image
width (int | None): Target width in pixels (default: None)
height (int | None): Target height in pixels (default: None)
dst_buffer (Image | None): Optional pre-allocated buffer (default: None)

Note: At least one of width or height must be specified.

Returns: New Image at target size (or None if dst_buffer provided)

Buffer Reuse

For batch processing, reuse the destination buffer. The buffer allows dynamic resizing as long as it has enough capacity.

Example:

from pyimagecuda import Image, load, Resize, save

# Pre-allocate destination with max capacity (e.g., 4K)
dst = Image(3840, 2160)

for file in image_files:
    src = load(file)

    # dst automatically adjusts its dimensions to the result
    # No need to calculate height or resize dst manually
    Resize.lanczos(src, width=800, dst_buffer=dst)

    save(dst, f"resized_{file}")
    src.free()

dst.free()

When using dst_buffer, the buffer's capacity must be large enough to hold the result. The buffer's logical dimensions (width, height) are automatically updated to match the operation result.