Showcase Workflows

This page describes the workflows FiLark is meant to support. It is not a benchmark report and does not duplicate the paper.

Video Overview

• Video S1 (Bilibili backup): interactive exploration, ROI analysis, and full-data processing.

• Video S2 (Bilibili backup): large-directory browsing, display filtering, auto-scroll, and annotation.

• Video S3 (Bilibili backup): simulated real-time monitoring, overlay results, and manual recheck.

Browse A Large Recording

1. Open a file or folder-backed source.

2. Use F and Y to fit many channels into view.

3. Use X to compress time when scanning long intervals.

4. Pan or run realtime auto-scroll.

5. Return to local scale when a candidate event appears.

The viewer keeps memory tied to the display buffer, so the same workflow applies to short files and long stitched recordings.

Inspect A Region Of Interest

1. Open the Algorithms tab.

2. Enable ROI mode and draw a region.

3. Open PSD, STFT, F-K, tau-p, RMS, SVD, xcorr, or denoising views.

4. Tune parameters locally before deciding whether to export or run a full source workflow.

This keeps the main canvas focused on context while heavier analysis lives in a separate window.

Create Labels While Browsing

1. Open the Annotation tab.

2. Choose bbox, polyline, or mask brush.

3. Draw labels in source coordinates.

4. Use label styles and visibility to manage dense sessions.

5. Export annotation JSON.

The exported file contains both labels and source metadata, so later scripts can recover the original data region.

Review Detector Output

1. Register a MonitorPlugin through GuiConfig.

2. Run it on the current view, full source, or realtime display.

3. Save monitor output as annotation JSON.

4. Import the JSON in the Annotation tab.

5. Correct bbox, polyline, or mask labels manually.

This lets algorithm results become reviewed labels without mixing temporary detection overlays with persistent annotation state.

Use FiLark Without The GUI

1. Open a Tape from filark.io.

2. Build an array transform chain from filark.dsp.transforms.

3. Execute in memory for small arrays or use filark.pipeline for large sources.

4. Reuse annotation JSON with filark.utils.annotations to extract labeled regions.

The same source metadata and coordinate conventions are used by the GUI and headless workflows.