Goal

Enable selection of samples in the HyperView interface, and provide an option to export the selected samples as a Torch-compatible dataset. This will support both prototyping and downstream ML workflows.

Requirements

• User should be able to select samples from the data view.
• Provide export functionality (button/menu) for selected samples.
• Exported dataset should be in a format readily usable by PyTorch (torch.utils.data.Dataset).
• Document the data schema and any requirements for serialization (e.g., images, labels, metadata).
• Ensure compatibility with common ML data loading operations (e.g., batching, transforms).
• Example usage should be part of the documentation.

Suggested Implementation Steps

1. Add a selection mechanism to the sample view (e.g., checkboxes, multi-select).
2. Implement an export option in the UI for the selected samples.
3. On export, package the selected samples into a Torch-compatible dataset object, and serialize it (e.g., as .pt or a folder structure).
4. Provide sample code (Python) for loading/exporting the dataset and for a minimal training loop using the exported dataset.

Documentation

• Add documentation to the repo on the selection/export workflow.
• Include code snippets for using the exported dataset in PyTorch.

Acceptance Criteria

• Users can select samples and export them as a Torch dataset.
• Exported dataset is loadable using PyTorch with correct metadata.
• Documentation and example code are available.