Rendering

A localization list is not an image — the rendering step turns the BasicSMLD into a super-resolution picture by accumulating every emitter onto a finely subdivided pixel grid. It is selected by a RenderConfig and backed by SMLMRender, which offers several rendering strategies, intensity- or field-based coloring, and PNG export.

analyze(smld, RenderConfig(zoom = 20, colormap = :inferno))   # → (smld, StepInfo)

Rendering is a pass-through: it writes the image to disk and returns the input smld unchanged. That is deliberate — a render can sit anywhere in the pipeline and be repeated at different zooms or colormaps without altering the data flowing to later steps. The pipeline treats it as a non-SMLD-producing step.

When to use / prerequisites

• Run on a BasicSMLD of localizations — typically as the final step, after detection/fitting, a quality filter, frame connection, and drift correction.
• Because it does not consume or transform the data, it is freely repeatable: drop in several RenderConfig steps (different strategies, zoom levels, colormaps, or color_by fields) and each writes its own image.
• To capture the image array in code (rather than only on disk), call SMLMRender.render(smld, cfg) or the step helper render_step(smld, cfg) directly — both return (image, RenderInfo).

Inputs, returns & artifacts

• Input: the current smld.
• Returns: (smld, StepInfo) — the same smld passes through. The RenderInfo (output size, pixel size, strategy, color mode) is on StepInfo.info.
• Artifacts (when outdir is set): a PNG written into the step folder with a descriptive name built from the run — <strategy>_<colormap>_<color_by>_<zoom>x.png (e.g. gaussianrender_inferno_20x.png); config.toml and info.toml; and at STANDARD verbosity a stats.md. No checkpoint is written (nothing changes). If you set RenderConfig(filename = ...) yourself, that path is used instead of the auto-generated name.

Concept

Each localization is placed on an output grid finer than the camera pixels. The strategy decides what gets drawn at each position; the coloring decides what value maps to color. See the SMLMRender documentation for the details of each.

• Strategies (strategy): HistogramRender() (fast binning), GaussianRender() (smooth blobs, the default), CircleRender() (anti-aliased circles at localization precision), and EllipseRender().
• Coloring: intensity-based — accumulate counts and apply colormap (:inferno, :hot, :magma); or field-based — set color_by to an emitter field (:z, :photons, :frame, :σ_x) and color by that value with a perceptual map like :turbo or :viridis. CircleRender/EllipseRender require color_by or a manual color.
• Resolution: zoom renders the exact camera FOV with subdivided pixels (zoom=20 on a 128×128 camera → 2560×2560), giving predictable, reproducible sizes; pixel_size (in nm) instead uses the data bounds with a margin. roi (zoom mode) renders a subset of the FOV as camera pixel ranges.

Configuration

RenderConfig is re-exported from SMLMRender (the "upstream owns the config" idiom). Its fields map 1:1 to render() keyword arguments. The ones you set most:

See the SMLMRender documentation for the full field list (target, color, field_range, field_clip_percentiles, scalebar_color, …) and defaults.

# Two renders of the same SMLD — both pass smld through unchanged
(smld, _) = analyze(smld, RenderConfig(zoom = 20, colormap = :inferno, scalebar = true))
(smld, _) = analyze(smld, RenderConfig(zoom = 10, color_by = :photons, colormap = :viridis))

# Get the image array directly
(img, rinfo) = SMLMRender.render(smld, RenderConfig(zoom = 20))

Output & interpretation

The step's StepInfo.summary reports:

stats.md additionally records the output pixel size (nm), color mode, and render time. Sanity checks: output_size should equal zoom × the camera dimensions (within the roi) in zoom mode; n_locs should match the input localization count unless some fell outside the rendered FOV. If a histogram render looks washed out, lower clip_percentile (or set it to nothing for saturate mode); for field coloring, confirm the chosen color_by field exists on the emitters.

Notes & caveats

• Pass-through, not a transform. The returned smld is the input; later steps see the data exactly as before. Position the render anywhere and repeat it freely.
• No checkpoint. Because nothing changes, no SMLD checkpoint is written for this step.
• Coordinates are in microns; zoom/roi are expressed in camera pixels, pixel_size/scalebar_length in nm/µm respectively.
• GPU rendering is available via backend = :cuda / :metal for large outputs.

References

No external citation — super-resolution histogram/Gaussian rendering is a standard SMLM visualization. See the SMLMRender documentation for the rendering strategies, coloring modes, and all configuration options.