Multimodal CIFAR Dataset

Quick Start

from cifar_rgb.datasets import create_cifar_rgb_loaders
from cifar_rgb.transforms import grayscale

train_loader, test_loader = create_cifar_rgb_loaders(
    batch_size=32,
    transform=grayscale,
    train_ratio=0.8,
    random_seed=42
)

for data, target in train_loader:
    pass  # data shape: (32, 3, 32, 32), target shape: (32,)

Visualization

import matplotlib.pyplot as plt
from cifar_rgb.datasets import create_cifar_rgb_loaders
from cifar_rgb.transforms import grayscale

train_loader, _ = create_cifar_rgb_loaders(
    batch_size=4,
    transform=grayscale,
    train_ratio=0.8,
    random_seed=42
)

img_rgb, label = next(iter(train_loader))[0][0], next(iter(train_loader))[1][0].item()

plt.figure(figsize=(10, 3))
plt.subplot(1, 4, 1)
plt.imshow(img_rgb.permute(1, 2, 0).numpy())
plt.title(f"RGB (Label: {label})")
plt.axis('off')

for i, (channel, title) in enumerate(zip(img_rgb, ["R/Mode 1", "G/Mode 2", "B/Mode 3"]), start=2):
    plt.subplot(1, 4, i)
    plt.imshow(channel.numpy(), cmap='gray')
    plt.title(title)
    plt.axis('off')

plt.tight_layout()
plt.show()

API Reference

RGBDataset

Wraps three grayscale datasets as RGB.

from cifar_rgb.datasets import RGBDataset

dataset = RGBDataset(
    mode1_dataset,
    mode2_dataset,
    mode3_dataset,
    train=True,
    train_ratio=0.8,
    random_seed=42
)

CIFARRGBData

High-level manager for CIFAR RGB datasets.

from cifar_rgb.datasets import CIFARRGBData

manager = CIFARRGBData(
    data_root='./data',
    transform=grayscale,
    train_ratio=0.8,
    random_seed=42
)

train_dataset, test_dataset = manager.get_datasets()
train_loader, test_loader = manager.get_loaders(
    batch_size=32,
    shuffle_train=True,
    shuffle_test=False,
    num_workers=0
)

Helper Functions

from cifar_rgb.datasets import create_cifar_rgb_datasets, create_cifar_rgb_loaders

# Datasets
train_dataset, test_dataset = create_cifar_rgb_datasets(
    data_root='./data',
    transform=grayscale,
    train_ratio=0.8,
    random_seed=42,
    download=True
)

# Loaders
train_loader, test_loader = create_cifar_rgb_loaders(
    batch_size=32,
    data_root='./data',
    transform=grayscale,
    train_ratio=0.8,
    random_seed=42,
    shuffle_train=True,
    shuffle_test=False,
    num_workers=0,
    download=True
)

Transforms

from cifar_rgb.transforms import grayscale
# grayscale = Compose([Grayscale(num_output_channels=1), ToTensor()])

Data Format

• Input: Three grayscale datasets
• Output: RGB images with:
  • Red = mode1
  • Green = mode2
  • Blue = mode3
• Shape: (3, 32, 32)
• Labels: Taken from original datasets

Dependencies

• PyTorch
• torchvision
• matplotlib
• pytest (for tests)

Testing

pytest tests/

Examples

See examples/visualize.py for RGB channel visualization.