Skip to content

Picker

xopr_viewer.picker.GroundingLinePicker

Bases: Parameterized

Interactive point picker for echogram annotation.

Click on the image to add points. Points are always stored in canonical coordinates (slow_time, twtt) regardless of the current display axis mode, and converted to/from display coordinates at the boundary.

Parameters:

  • image (Image) –

    The echogram image to annotate.

  • ds (Dataset, default: None ) –

    The echogram dataset (needed for coordinate conversion when axis modes other than the default are used).

  • x_mode (str, default: 'gps_time' ) –

    X-axis display mode (default: "gps_time").

  • y_mode (str, default: 'twtt' ) –

    Y-axis display mode (default: "twtt").

  • layers (dict, default: None ) –

    Dictionary of layer name -> xr.Dataset for snap-to-layer.

  • snap_threshold (float, default: 5.0 ) –

    Snap threshold in microseconds (default: 5.0).

df property 

df: DataFrame

Get picked points as a DataFrame (canonical coordinates).

visible_layers property writable 

visible_layers: list[str] | None

Layer names to snap to, or None for all layers.

clear 

clear() -> None

Clear all picked points.

delete_by_id 

delete_by_id(point_id: str) -> None

Delete a specific point by ID.

element 

element(width: int = 700, height: int = 400) -> Overlay

Get the HoloViews element (image + points overlay).

from_csv 

from_csv(path: str) -> None

Load picked points from CSV (expects canonical coordinates).

panel 

panel(width: int = 700, height: int = 400) -> Column

Get a Panel layout with the picker and control buttons.

set_axis_modes 

set_axis_modes(x_mode: str, y_mode: str, *, _notify: bool = True) -> None

Update current axis modes (called when sidebar dropdowns change).

Parameters:

  • _notify (bool, default: True ) –

    If True (default), send on the points pipe to re-render points in the new coordinate system. Pass False when calling from inside a DynamicMap callback to avoid recursive overlay resolution.

to_csv 

to_csv(path: str) -> None

Export picked points to CSV (canonical coordinates).

undo 

undo() -> None

Remove the last added point.

xopr_viewer.picker.pick_echogram 

pick_echogram(
    data: DataArray | Image,
    x_dim: str = "trace",
    y_dim: str = "twtt_us",
    **opts,
) -> GroundingLinePicker

Convenience function to create a picker from xarray or holoviews data.

Parameters:

  • data (DataArray or Image) –

    The echogram data

  • x_dim (str, default: 'trace' ) –

    X dimension name

  • y_dim (str, default: 'twtt_us' ) –

    Y dimension name

  • **opts –

    Additional options passed to hv.Image.opts()

Returns:

xopr_viewer.picker.compute_layer_slope 

compute_layer_slope(twtt: DataArray, smoothing_window: int = 1) -> DataArray

Compute slope of a layer profile after smoothing.

Smoothing uses xr.DataArray.rolling with a centered uniform (box-car) window followed by .mean(). Every output point is the unweighted mean of the smoothing_window nearest input points. The window is centered so the smoothed value at index i averages indices i - window//2 … i + window//2.

Points within window // 2 of either edge have fewer neighbors, so rolling().mean() returns NaN there. These NaN edges propagate through the derivative and appear as gaps at the ends of the slope curve.

The derivative is computed by xr.DataArray.differentiate, which uses second-order central finite differences in the interior and first-order one-sided differences at the boundaries. Because it operates on coordinate values (not integer indices), the result is in physical units: d(twtt) / d(coordinate). For datetime coordinates like slow_time the denominator is in nanoseconds (NumPy's internal representation), giving slope in seconds per nanosecond.

Developer notes — common customizations:

  • Weighted smoothing — Replace rolling().mean() with a Gaussian kernel (scipy.ndimage.gaussian_filter1d) or a Savitzky–Golay filter (scipy.signal.savgol_filter) to better preserve sharp features. Both require scipy.
  • Edge handling — Pass min_periods=1 to rolling() to get biased shorter-window estimates near boundaries instead of NaN gaps.
  • Units — For datetime coordinates, multiply by 1e9 to convert from seconds/nanosecond to dimensionless seconds/second, or convert the coordinate to elapsed seconds before calling this function.

Parameters:

  • twtt (DataArray) –

    Two-way travel time values in seconds, indexed by a single dimension.

  • smoothing_window (int, default: 1 ) –

    Size of the rolling mean window for smoothing (1 = no smoothing).

Returns:

  • DataArray –

    Slope as d(twtt)/d(coordinate) along the first dimension.