Computer Vision Datasets

xTorch provides an extensive collection of built-in dataset handlers for a wide variety of computer vision tasks, from image classification and object detection to semantic segmentation and beyond. This allows you to easily benchmark models on standard academic datasets without writing custom data loading code.

All computer vision datasets are located under the xt::datasets namespace and can be found within the <xtorch/datasets/computer_vision/> header directory.

General Usage

The standard workflow for using any computer vision dataset involves defining a pipeline of image transformations, instantiating the desired dataset class, and then passing it to a data loader.

#include <xtorch/xtorch.h>
 
int main() {
    // 1. Define a pipeline of image transformations for data augmentation
    auto transforms = std::make_unique<xt::transforms::Compose>(
        std::make_shared<xt::transforms::image::RandomHorizontalFlip>(),
        std::make_shared<xt::transforms::image::Resize>(std::vector<int64_t>{32, 32}),
        std::make_shared<xt::transforms::general::Normalize>(
            std::vector<float>{0.5, 0.5, 0.5},
            std::vector<float>{0.5, 0.5, 0.5}
        )
    );
 
    // 2. Instantiate a dataset for CIFAR-10
    auto dataset = xt::datasets::CIFAR10(
        "./data",
        xt::datasets::DataMode::TRAIN,
        /*download=*/true,
        std::move(transforms)
    );
 
    std::cout << "CIFAR-10 dataset size: " << *dataset.size() << std::endl;
 
    // 3. Pass the dataset to a DataLoader
    xt::dataloaders::ExtendedDataLoader data_loader(dataset, 128, true, 4);
 
    // The data loader is now ready for use in a training loop
    for (auto& batch : data_loader) {
        auto images = batch.first;
        auto labels = batch.second;
        // ... training step ...
    }
}

!!! info "Standard Dataset Constructors" Most dataset constructors follow a standard pattern: DatasetName(const std::string& root, DataMode mode, bool download, TransformPtr transforms)

• root: The directory where the data is stored or will be downloaded.
• mode: DataMode::TRAIN, DataMode::TEST, or DataMode::VALIDATION.
• download: If true, the dataset will be downloaded if not found in the root directory.
• transforms: A unique_ptr to a transform pipeline to be applied to the data.

Available Datasets by Task

Image Classification

Object Detection

Semantic & Instance Segmentation

Image Generation

Image Captioning

Autonomous Driving & 3D Vision

Optical Flow