Implementation:Junyanz Pytorch CycleGAN and pix2pix Combine A and B

**Metadata**
Knowledge Sources	pytorch-CycleGAN-and-pix2pix
Domains	Image-to-Image Translation, Data Preparation, Paired Image Translation
Last Updated	2026-02-09

Overview

The datasets/combine_A_and_B.py script is an external tool that concatenates corresponding image pairs from two separate directories into single side-by-side images for pix2pix training. It uses multiprocessing for efficient parallel processing of large datasets.

Description

The script iterates over phase directories (train, test, val) within the specified source folders. For each phase, it finds all images in the A directory, locates the corresponding image in the B directory (by matching filename), and concatenates them horizontally. The combined images are saved to an output directory preserving the phase structure.

The script supports two modes:

Default mode (--use_AB not set): Takes separate --fold_A and --fold_B directories
AB mode (--use_AB set): Assumes images are already in A and B subdirectories within the same parent

Multiprocessing is enabled by default and can be disabled with --no_multiprocessing for debugging.

Usage

Run as a command-line tool before pix2pix training when preparing custom paired datasets.

Code Reference

Source Location

File	Lines
datasets/combine_A_and_B.py	L1-68

Signature

# Command-line arguments
parser.add_argument("--fold_A", help="Input directory for domain A images",
                    type=str, required=True)
parser.add_argument("--fold_B", help="Input directory for domain B images",
                    type=str, required=True)
parser.add_argument("--fold_AB", help="Output directory for combined AB images",
                    type=str, required=True)
parser.add_argument("--num_imgs", help="Max number of images to process",
                    type=int, default=1000000)
parser.add_argument("--use_AB", help="If set, use A/B subdirectory structure",
                    action='store_true')
parser.add_argument("--no_multiprocessing",
                    help="Disable multiprocessing for debugging",
                    action='store_true')

Import

N/A -- This is a standalone command-line script, not an importable Python module.

I/O Contract

**Inputs**
Parameter	Type	Description
--fold_A	str (CLI arg)	Path to directory containing domain A images (with train/, test/, val/ subdirectories)
--fold_B	str (CLI arg)	Path to directory containing domain B images (matching structure)
--fold_AB	str (CLI arg)	Path to output directory for combined images
--num_imgs	int (CLI arg)	Maximum number of images to process per phase (default: 1000000)
--use_AB	flag (CLI arg)	If set, expects A/ and B/ subdirectories within source
--no_multiprocessing	flag (CLI arg)	If set, disables parallel processing

**Outputs**
Output	Type	Description
<fold_AB>/train/	directory	Combined AB training images (horizontally concatenated pairs)
<fold_AB>/test/	directory	Combined AB test images
<fold_AB>/val/	directory	Combined AB validation images (if val/ phase exists in source)

Usage Examples

# Combine separate A and B directories into pix2pix format
python datasets/combine_A_and_B.py \
    --fold_A ./datasets/my_dataset/A \
    --fold_B ./datasets/my_dataset/B \
    --fold_AB ./datasets/my_dataset

# Limit to first 100 images per phase
python datasets/combine_A_and_B.py \
    --fold_A ./datasets/my_dataset/A \
    --fold_B ./datasets/my_dataset/B \
    --fold_AB ./datasets/my_dataset \
    --num_imgs 100

# Disable multiprocessing for debugging
python datasets/combine_A_and_B.py \
    --fold_A ./datasets/my_dataset/A \
    --fold_B ./datasets/my_dataset/B \
    --fold_AB ./datasets/my_dataset \
    --no_multiprocessing

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment