Copy Make Border

Overview

The Copy Make Border sample demonstrates how to pad an image with a border of configurable width and fill style using CV-CUDA’s GPU-accelerated copymakeborder operator. The operator supports multiple border modes (CONSTANT, REPLICATE, REFLECT, REFLECT101, WRAP) and operates directly on device tensors, making it well-suited for preprocessing pipelines that need to pad images before inference.

Usage

Basic Usage

Add an orange border to the default cat image:

python3 copymakeborder.py

Custom Output Path

Specify a custom output file:

python3 copymakeborder.py -i input.jpg -o cat_copymakeborder.jpg

Command-Line Arguments

Argument	Short Form	Default	Description
`--input`	`-i`	tabby_tiger_cat.jpg	Input image file path
`--output`	`-o`	cvcuda/.cache/cat_copymakeborder.jpg	Output image file path

Implementation

Adding a Colored Border

# Add a visible border around the image using CONSTANT mode so the border
# is filled with a solid color rather than replicated/reflected pixels.
# The border widths (top=30, bottom=30, left=60, right=60) make the
# added region clearly visible in the output.
output_image: cvcuda.Tensor = cvcuda.copymakeborder(
    src=input_image,
    top=30,
    bottom=30,
    left=60,
    right=60,
    border_mode=cvcuda.Border.CONSTANT,
    # Bright orange border (R=255, G=140, B=0) makes the padding conspicuous.
    border_value=[255, 140, 0],
)
write_image(output_image, args.output)

Key points:

Border mode: cvcuda.Border.CONSTANT fills the added region with a fixed color; other modes (REPLICATE, REFLECT, REFLECT101, WRAP) derive fill values from existing image pixels instead.
Border value: A three-element list [R, G, B] supplies the fill color for CONSTANT mode; it is silently ignored for the other modes.
Output shape: The output tensor is automatically sized to (H + top + bottom, W + left + right, C); no pre-allocation is required.
Asymmetric padding: top, bottom, left, and right are independent integers, enabling padding that differs on each side — useful for letterboxing.

Expected Output

The output shows the original image surrounded by a 30-pixel vertical and 60-pixel horizontal orange border:

../../_images/tabby_tiger_cat.jpg — Original Input Image

../../_images/cat_copymakeborder.jpg — Output: Image with orange border padding

CV-CUDA Operators Used

Operator	Purpose
`cvcuda.copymakeborder()`	Pad an image with a configurable border width and fill mode

Common Utilities Used

read_image() - Load image as CV-CUDA tensor
write_image() - Save bordered image