Basic Usage¶

This page is a task-oriented walkthrough for a typical FiLark session. The demo videos provide longer examples, while the sections below keep individual workflows easy to find.

Demo Videos¶

Video	Workflow	Links
S1	Interactive exploration, ROI analysis, and full-data processing	YouTube / Bilibili backup
S2	Large-directory browsing, real-time display filtering, auto-scroll, and annotation	YouTube / Bilibili backup
S3	Simulated real-time monitoring, overlay results, and manual recheck	YouTube / Bilibili backup

1. Load A Source¶

Open a single H5/NPY/binary source or a folder-backed stream. The Load Data panel should show the source metadata that matters for DAS interpretation: layout, sample rate, channel spacing, dtype, dimensions, start time, and optional gauge length.

The key point for the viewer is that the source is treated as a continuous Tape, not as a pre-extracted snippet.

2. Navigate The Stream¶

Use mouse and keyboard navigation to move through time and channels:

pan horizontally and vertically;
zoom around the cursor;
fit the visible channel range with F;
adjust time/channel scale with X, Y, Shift+X, and Shift+Y;
jump directly to a time or sample index with Goto.

Videos S1 and S2 show the fixed-size GPU buffer as an interaction model: the viewport changes, but the viewer does not need to upload the entire dataset.

3. Adjust The Display¶

The display is allowed to change without mutating the source data. A concise recording should show colormap changes, color-limit adjustment, real-time display filtering, and scale changes.

Note

Display filters are viewer-side tools. They are useful for exploration, but they do not rewrite the underlying Tape source.

4. Inspect An ROI¶

Select an ROI and open diagnostic views such as PSD, STFT, F-K, tau-p, RMS, SVD, cross-correlation, or denoising previews.

The ROI step is the bridge between interactive visual browsing and processing: it lets a user test a transform on a small region before committing to the full source.

5. Process And Compare¶

After an ROI operation looks reasonable, submit the same operation to the full dataset. FiLark can open the original and processed results in linked tabs so that view positions remain synchronized during comparison.

Video S1 shows an ROI-tested operation applied to the full source and then compared in linked tabs.

6. Annotate Events¶

Use the Annotation panel to draw:

bbox labels for compact events;
polyline labels for curving or elongated patterns;
mask labels for pixel-level regions.

Mask brush width is screen-pixel based while saved masks are stored in data coordinates. Video S2 shows bbox and polyline annotation, export, reload, and review.

7. Run A Monitor¶

Open a monitor plugin from the GUI, choose its scope and output path, and run it while the viewer remains interactive. Monitor results should appear as ordinary overlay events: bbox, polyline, and mask.

For a demo recording, use a deterministic source and configuration. The monitor does not need to be the final research model; the clip should demonstrate that algorithm results can arrive asynchronously without blocking navigation.

8. Recheck Saved Results¶

Saved monitor output should be reloadable through the annotation module. This is important because detector output is rarely final: a user should be able to review, adjust, delete, or add labels after inference.

Video S3 shows monitor output saved, reopened, and reviewed through the annotation workflow.