Skip to main content

Datasets Module

The tryon.datasets module provides easy-to-use interfaces for downloading and loading datasets commonly used in fashion and virtual try-on applications.

Overview​

The datasets module uses a class-based adapter pattern where each dataset is implemented as a class that extends the base Dataset interface. This design ensures consistency across different datasets while allowing for dataset-specific features.

Key Features​

  • Consistent API: All datasets follow the same interface pattern
  • Automatic Download: Built-in download functionality for supported datasets
  • Memory Efficient: Support for lazy loading with PyTorch DataLoader
  • Easy to Extend: Simple interface for adding new datasets
  • State Management: Automatic caching and metadata tracking

Supported Datasets​

Fashion-MNIST​

A dataset of Zalando's article images designed as a drop-in replacement for the original MNIST dataset. Ideal for quick prototyping, learning, and benchmarking.

  • 60,000 training examples
  • 10,000 test examples
  • 10 classes: T-shirt/top, Trouser, Pullover, Dress, Coat, Sandal, Shirt, Sneaker, Bag, Ankle boot
  • 28×28 grayscale images
  • ~60MB total size

Learn more about Fashion-MNIST →

VITON-HD​

A high-resolution virtual try-on dataset consisting of person images and clothing images. Perfect for training and evaluating virtual try-on models.

  • 11,647 training pairs
  • 2,032 test pairs
  • 1024×768 resolution images
  • Person and clothing image pairs

Learn more about VITON-HD →

Subjects200K​

A large-scale dataset containing 200,000 paired images for subject consistency research. Each image pair maintains subject consistency while presenting variations in scene context. Loaded from HuggingFace.

  • ~200,000 paired images
  • Three collections: Collection 1 (512×512), Collection 2 (512×512), Collection 3 (1024×1024)
  • Quality assessment scores for filtering
  • HuggingFace integration (automatic download)

Learn more about Subjects200K →

Quick Start​

Fashion-MNIST​

from tryon.datasets import FashionMNIST

# Create dataset instance (downloads automatically)
dataset = FashionMNIST(download=True)

# Load the dataset
(train_images, train_labels), (test_images, test_labels) = dataset.load(
    normalize=True,
    flatten=False
)

print(f"Training set: {train_images.shape}")  # (60000, 28, 28)
print(f"Test set: {test_images.shape}")      # (10000, 28, 28)
print(f"Class 0: {dataset.get_class_name(0)}")  # 'T-shirt/top'

VITON-HD​

from tryon.datasets import VITONHD
from torchvision import transforms

# Create dataset instance
dataset = VITONHD(data_dir="./datasets/viton_hd", download=False)

# Define transforms
transform = transforms.Compose([
    transforms.Resize((512, 384)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
])

# Get DataLoader for efficient lazy loading
train_loader = dataset.get_dataloader(
    split='train',
    batch_size=8,
    shuffle=True,
    transform=transform
)

# Use in training loop
for batch in train_loader:
    person_imgs = batch['person']  # [batch_size, 3, H, W]
    clothing_imgs = batch['clothing']  # [batch_size, 3, H, W]
    # Train model...

Subjects200K​

from tryon.datasets import Subjects200K
from torchvision import transforms

# Create dataset instance (loads from HuggingFace)
dataset = Subjects200K()

# Get HuggingFace dataset
hf_dataset = dataset.get_hf_dataset()
sample = hf_dataset['train'][0]
image = sample['image']  # PIL Image (composite with paired images)
collection = sample['collection']  # 'collection_1', 'collection_2', or 'collection_3'

# Get PyTorch DataLoader with quality filtering
transform = transforms.Compose([
    transforms.Resize((512, 512)),
    transforms.ToTensor(),
])

dataloader = dataset.get_dataloader(
    batch_size=16,
    transform=transform,
    collection='collection_2',
    filter_high_quality=True
)

# Use in training loop
for batch in dataloader:
    images = batch['image']  # [batch_size, 3, H, W]
    collections = batch['collection']
    quality_assessments = batch['quality_assessment']
    # Train model...

Architecture​

All dataset classes extend the Dataset base class, which provides:

  • Automatic download management: Handles dataset downloads and verification
  • State tracking: Tracks normalization, flattening, and other transformations
  • Metadata management: Stores dataset information and statistics
  • Consistent interface: Common methods like load(), get_info(), get_class_names()

Base Class Interface​

from tryon.datasets import Dataset
from pathlib import Path

class MyDataset(Dataset):
    def _get_default_data_dir(self) -> Path:
        """Return default data directory."""
        return Path.home() / '.opentryon' / 'datasets' / 'my_dataset'
    
    def _ensure_downloaded(self) -> None:
        """Download dataset files if needed."""
        # Implement download logic
        pass
    
    def load(self, **kwargs):
        """Load and return dataset."""
        # Implement loading logic
        return (train_data, train_labels), (test_data, test_labels)
    
    def get_class_names(self) -> list:
        """Return list of class names."""
        return ['class1', 'class2', ...]

Dataset Storage​

By default, datasets are stored in:

  • Linux/Mac: ~/.opentryon/datasets/
  • Windows: C:\Users\<username>\.opentryon\datasets\

You can override this by specifying a custom data_dir when creating a dataset instance.

Requirements​

Fashion-MNIST​

  • No additional requirements: Uses standard Python libraries

VITON-HD​

  • PyTorch: Required for DataLoader support
  • torchvision: Required for transforms
  • Pillow: Required for image loading

Install with:

pip install torch torchvision pillow

Subjects200K​

  • datasets: HuggingFace datasets library (required)
  • PyTorch: Required for DataLoader support
  • torchvision: Required for transforms
  • Pillow: Required for image loading

Install with:

pip install datasets torch torchvision pillow

Next Steps​