The Library
The Library is RIA Hub’s central catalogue of all assets stored across your repositories. Any file committed to a repository that RIA Hub recognises — a recording, a dataset, a trained model — is automatically indexed and made searchable here, with metadata extracted and thumbnails generated without any extra steps from you.
The Library is where you:
- Find recordings captured by a Campaign Control run
- Browse datasets generated or curated in the Dataset Manager
- Locate trained model files (.onnx, .pt, .pth) ready to load into the Zone Fingerprinting Demo
- Quickly inspect a recording’s spectrogram, constellation, or PSD before deciding whether to use it
- Move assets between repositories using the commit-to-repo workflow
What you’ll need
Section titled “What you’ll need”- At least one repository containing indexed files — files are indexed automatically when committed via RIA Hub’s upload interface or the Conductor
- For locally-managed repos: Git LFS must be set up so that large binary files are stored correctly (see Working with Git LFS)
Navigating the Library
Section titled “Navigating the Library”Click Library in the top navigation. The Library is organised into tabs by asset type:
| Tab | What it contains |
|---|---|
| Recordings | SigMF recordings (.sigmf-data/.sigmf-meta pairs), NumPy files, WAV files |
| Radio Datasets | HDF5 datasets, CSV files, Parquet files |
| PyTorch State Dicts | Trained model weights (.pt, .pth) |
| ONNX Graphs | Exported inference models (.onnx) |
| PyTorch Modules | Module definitions |
| Model Builder Tasks | ML pipeline definitions |
| Holoscan Inference Apps | Inference application specs |
| Actions | Workflow and action definitions |
Select the tab for the asset type you’re looking for.
Filtering and searching
Section titled “Filtering and searching”Each tab shares the same set of filters:
- Repository — limit results to one or more specific repositories
- Directory — filter by folder path within a repository
- Branch — filter by git branch
Use the filter dropdowns to add criteria. Active filters appear as tags above the table and can be removed individually or all at once with Clear All Filters. The text search box matches across all visible columns including metadata fields.
Metadata columns are dynamic — each asset type exposes different fields (e.g. sample_rate, center_frequency, num_classes). Use the column visibility popover to show or hide the fields that matter for your current task.
Inspecting a recording
Section titled “Inspecting a recording”For Recordings, every row shows a spectrogram thumbnail. Click the thumbnail (or anywhere on the row) to open the Quick View panel with seven visualisation tabs:
| Tab | What it shows |
|---|---|
| Spectrogram | Time vs. frequency power — the fastest way to assess signal presence and quality |
| Constellation | IQ scatter plot — reveals modulation shape and phase errors |
| PSD | Power Spectral Density — frequency distribution of energy |
| Time Series | Raw I and Q amplitude over time |
| FFT | Single-frame frequency view |
| Frequency Spectrum | Frequency spectrum plot |
| 3D Spectrogram | Depth-enhanced time-frequency view |
The panel also shows the recording’s sample count, data type, and sample rate.
Downloading assets
Section titled “Downloading assets”Click the Download button on any row to download the file directly. For LFS-tracked files the download is served through RIA Hub’s LFS backend — the raw content is downloaded, not the pointer file.
To download multiple files, check the boxes next to the rows you want and click Download All Selected.
Moving assets between repositories
Section titled “Moving assets between repositories”Any LFS-tracked asset can be copied into another repository without re-uploading the file content — RIA Hub links the existing LFS object to the new repository path.
- Select one or more rows using the checkboxes
- Click Commit Selected to Repo
- Choose the target repository (search by name or
owner/repo) - Select the target branch, or leave it to commit to the default branch
- Enter a commit message
- Click Commit
The LFS pointer is created in the target repository, pointing at the same underlying content. The file appears in the target repo immediately and is re-indexed in the Library under its new location.
Next steps
Section titled “Next steps”- Review and label recordings — Reviewing and Labelling Recordings explains how to inspect quality and add annotations before curating.
- Build a dataset — Curating a Dataset turns labelled recordings into a training-ready HDF5 dataset.
- Upload large files — Working with Git LFS covers file size limits and how to push assets from your local machine.