Center Crop

Overview

The Center Crop sample demonstrates symmetric center-cropping of an image using CV-CUDA’s GPU-accelerated center_crop operator. The operator extracts a rectangular region from the geometric centre of the image without requiring the caller to compute corner offsets manually.

Usage

Basic Usage

Crop the default cat image to 224×224 pixels:

python3 center_crop.py -i input.jpg

Custom Crop Size

Specify a different crop width and height:

python3 center_crop.py -i input.jpg -o cat_center_crop.jpg --width 320 --height 240

Command-Line Arguments

Argument	Short Form	Default	Description
--input	-i	tabby_tiger_cat.jpg	Input image file path
--output	-o	cvcuda/.cache/cat_center_crop.jpg	Output image file path
--width		224	Crop width in pixels (clamped to input width)
--height		224	Crop height in pixels (clamped to input height)

Implementation

Center Crop

# Derive a square crop size that fits within the image dimensions.
# The operator requires [crop_height, crop_width] as a Python list.
h, w = input_image.shape[0], input_image.shape[1]
crop_h = min(args.height, h)
crop_w = min(args.width, w)
crop_size = [crop_h, crop_w]

Key points:

1. No coordinate math: cvcuda.center_crop computes the top-left corner internally, so the caller only needs to supply the desired [height, width].

2. crop_size list: The second argument is a two-element Python list [crop_height, crop_width] — not a tuple.

3. Layout preserved: The output tensor shares the same layout (HWC/NHWC) and dtype as the input; no conversion is needed before writing.

4. Clamping: The sample clamps the requested crop dimensions to the actual image size to avoid an out-of-bounds error when the crop is larger than the source.

5. Single-image usage: The sample operates on an HWC tensor directly; batched NHWC usage follows the same crop_size argument convention.

Expected Output

The output is the center portion of the input image at the requested dimensions (default 224×224):

Original Input Image

Output: Center-cropped to 224×224

CV-CUDA Operators Used

Operator	Purpose
cvcuda.center_crop()	Symmetrically crop a rectangular region from the centre of an image

Common Utilities Used

• read_image() - Load image as CV-CUDA tensor

• write_image() - Save cropped image

See Also

• Resize Operator - Scale images to arbitrary dimensions

• Common Utilities - Helper functions used across samples