scitex 2.7.3__py3-none-any.whl → 2.10.0__py3-none-any.whl

scitex/fts/_fig/_editor/_gui/_flask_editor/_bbox.py

@@ -0,0 +1,1339 @@
+#!/usr/bin/env python3
+# File: ./src/scitex/vis/editor/flask_editor/bbox.py
+"""Bounding box extraction for figure elements.
+
+Updated to integrate Schema v0.3 geometry extraction for shape-based hit testing.
+"""
+
+from typing import Any, Dict
+
+# Try to import schema v0.3 geometry extraction
+try:
+    from scitex.plt.utils.metadata._geometry_extraction import (
+        extract_axes_bbox_px,
+        extract_bar_group_geometry,
+        extract_line_geometry,
+        extract_polygon_geometry,
+        extract_scatter_geometry,
+    )
+
+    GEOMETRY_V03_AVAILABLE = True
+except ImportError:
+    GEOMETRY_V03_AVAILABLE = False
+
+
+def extract_bboxes(
+    fig, ax, renderer, img_width: int, img_height: int
+) -> Dict[str, Any]:
+    """Extract bounding boxes for all figure elements (single-axis)."""
+    from matplotlib.transforms import Bbox
+
+    # Get figure tight bbox in inches
+    fig_bbox = fig.get_tightbbox(renderer)
+    tight_x0 = fig_bbox.x0
+    tight_y0 = fig_bbox.y0
+    tight_width = fig_bbox.width
+    tight_height = fig_bbox.height
+
+    # bbox_inches='tight' adds pad_inches (default 0.1) around the tight bbox
+    pad_inches = 0.1
+    saved_width_inches = tight_width + 2 * pad_inches
+    saved_height_inches = tight_height + 2 * pad_inches
+
+    # Scale factors for converting inches to pixels
+    scale_x = img_width / saved_width_inches
+    scale_y = img_height / saved_height_inches
+
+    bboxes = {}
+
+    def get_element_bbox(element, name):
+        """Get element bbox in image pixel coordinates."""
+        try:
+            bbox = element.get_window_extent(renderer)
+
+            elem_x0_inches = bbox.x0 / fig.dpi
+            elem_x1_inches = bbox.x1 / fig.dpi
+            elem_y0_inches = bbox.y0 / fig.dpi
+            elem_y1_inches = bbox.y1 / fig.dpi
+
+            x0_rel = elem_x0_inches - tight_x0 + pad_inches
+            x1_rel = elem_x1_inches - tight_x0 + pad_inches
+            y0_rel = saved_height_inches - (elem_y1_inches - tight_y0 + pad_inches)
+            y1_rel = saved_height_inches - (elem_y0_inches - tight_y0 + pad_inches)
+
+            bboxes[name] = {
+                "x0": max(0, int(x0_rel * scale_x)),
+                "y0": max(0, int(y0_rel * scale_y)),
+                "x1": min(img_width, int(x1_rel * scale_x)),
+                "y1": min(img_height, int(y1_rel * scale_y)),
+                "label": name.replace("_", " ").title(),
+            }
+        except Exception as e:
+            print(f"Error getting bbox for {name}: {e}")
+
+    def bbox_to_img_coords(bbox):
+        """Convert matplotlib bbox to image pixel coordinates."""
+        x0_inches = bbox.x0 / fig.dpi
+        y0_inches = bbox.y0 / fig.dpi
+        x1_inches = bbox.x1 / fig.dpi
+        y1_inches = bbox.y1 / fig.dpi
+        x0_rel = x0_inches - tight_x0 + pad_inches
+        y0_rel = y0_inches - tight_y0 + pad_inches
+        x1_rel = x1_inches - tight_x0 + pad_inches
+        y1_rel = y1_inches - tight_y0 + pad_inches
+        return {
+            "x0": int(x0_rel * scale_x),
+            "y0": int((saved_height_inches - y1_rel) * scale_y),
+            "x1": int(x1_rel * scale_x),
+            "y1": int((saved_height_inches - y0_rel) * scale_y),
+        }
+
+    # Get bboxes for title, labels
+    # Use ax_00_ prefix for consistency with geometry_px.json format
+    ax_prefix = "ax_00_"
+    if ax.title.get_text():
+        get_element_bbox(ax.title, f"{ax_prefix}title")
+    if ax.xaxis.label.get_text():
+        get_element_bbox(ax.xaxis.label, f"{ax_prefix}xlabel")
+    if ax.yaxis.label.get_text():
+        get_element_bbox(ax.yaxis.label, f"{ax_prefix}ylabel")
+
+    # Get axis bboxes
+    _extract_axis_bboxes(ax, renderer, bboxes, bbox_to_img_coords, Bbox, ax_prefix)
+
+    # Get legend bbox
+    legend = ax.get_legend()
+    if legend:
+        get_element_bbox(legend, f"{ax_prefix}legend")
+
+    # Get caption bbox (figure-level text)
+    for text_artist in fig.texts:
+        # Check if this is a caption (positioned below figure)
+        text_content = text_artist.get_text()
+        if text_content:
+            pos = text_artist.get_position()
+            # Caption is typically centered horizontally (x ~= 0.5) and below figure (y < 0.1)
+            if pos[1] < 0.1:
+                get_element_bbox(text_artist, f"{ax_prefix}caption")
+                break  # Only one caption expected
+
+    # Get trace (line) bboxes
+    _extract_trace_bboxes(
+        ax,
+        fig,
+        renderer,
+        bboxes,
+        get_element_bbox,
+        tight_x0,
+        tight_y0,
+        saved_height_inches,
+        scale_x,
+        scale_y,
+        pad_inches,
+    )
+
+    # Add schema v0.3 metadata if available
+    if GEOMETRY_V03_AVAILABLE:
+        axes_bbox = extract_axes_bbox_px(ax, fig)
+        bboxes["_meta"] = {
+            "schema_version": "0.3.0",
+            "axes_bbox_px": axes_bbox,
+            "geometry_available": True,
+        }
+
+    return bboxes
+
+
+def extract_bboxes_multi(
+    fig, axes_map: Dict[str, Any], renderer, img_width: int, img_height: int
+) -> Dict[str, Any]:
+    """Extract bounding boxes for all elements in a multi-axis figure.
+
+    Args:
+        fig: Matplotlib figure
+        axes_map: Dict mapping axis IDs (e.g., 'ax_00') to matplotlib Axes objects
+        renderer: Matplotlib renderer
+        img_width: Image width in pixels
+        img_height: Image height in pixels
+
+    Returns:
+        Dict with bboxes keyed by "{ax_id}_{element_type}" (e.g., "ax_00_xlabel")
+    """
+    from matplotlib.transforms import Bbox
+
+    # Get figure tight bbox in inches
+    fig_bbox = fig.get_tightbbox(renderer)
+    tight_x0 = fig_bbox.x0
+    tight_y0 = fig_bbox.y0
+    tight_width = fig_bbox.width
+    tight_height = fig_bbox.height
+
+    # bbox_inches='tight' adds pad_inches (default 0.1) around the tight bbox
+    pad_inches = 0.1
+    saved_width_inches = tight_width + 2 * pad_inches
+    saved_height_inches = tight_height + 2 * pad_inches
+
+    # Scale factors for converting inches to pixels
+    scale_x = img_width / saved_width_inches
+    scale_y = img_height / saved_height_inches
+
+    bboxes = {}
+
+    def get_element_bbox(element, name, ax_id, current_ax=None):
+        """Get element bbox in image pixel coordinates."""
+        import numpy as np
+
+        full_name = f"{ax_id}_{name}"
+        try:
+            bbox = element.get_window_extent(renderer)
+
+            # Check for invalid bbox (infinity or NaN)
+            if not (
+                np.isfinite(bbox.x0)
+                and np.isfinite(bbox.x1)
+                and np.isfinite(bbox.y0)
+                and np.isfinite(bbox.y1)
+            ):
+                # Try to get bbox from data for scatter/collection elements
+                if hasattr(element, "get_offsets") and current_ax is not None:
+                    offsets = element.get_offsets()
+                    if len(offsets) > 0 and np.isfinite(offsets).all():
+                        # Use axis transform to get display coordinates
+                        display_coords = current_ax.transData.transform(offsets)
+                        x0 = display_coords[:, 0].min()
+                        x1 = display_coords[:, 0].max()
+                        y0 = display_coords[:, 1].min()
+                        y1 = display_coords[:, 1].max()
+                        if np.isfinite([x0, x1, y0, y1]).all():
+                            from matplotlib.transforms import Bbox
+
+                            bbox = Bbox.from_extents(x0, y0, x1, y1)
+                        else:
+                            return  # Skip this element
+                    else:
+                        return  # Skip this element
+                else:
+                    return  # Skip this element
+
+            elem_x0_inches = bbox.x0 / fig.dpi
+            elem_x1_inches = bbox.x1 / fig.dpi
+            elem_y0_inches = bbox.y0 / fig.dpi
+            elem_y1_inches = bbox.y1 / fig.dpi
+
+            x0_rel = elem_x0_inches - tight_x0 + pad_inches
+            x1_rel = elem_x1_inches - tight_x0 + pad_inches
+            y0_rel = saved_height_inches - (elem_y1_inches - tight_y0 + pad_inches)
+            y1_rel = saved_height_inches - (elem_y0_inches - tight_y0 + pad_inches)
+
+            # Clamp values to avoid overflow
+            x0_px = max(0, min(img_width, int(x0_rel * scale_x)))
+            y0_px = max(0, min(img_height, int(y0_rel * scale_y)))
+            x1_px = max(0, min(img_width, int(x1_rel * scale_x)))
+            y1_px = max(0, min(img_height, int(y1_rel * scale_y)))
+
+            bboxes[full_name] = {
+                "x0": x0_px,
+                "y0": y0_px,
+                "x1": x1_px,
+                "y1": y1_px,
+                "label": f"{ax_id}: {name.replace('_', ' ').title()}",
+                "ax_id": ax_id,
+            }
+        except Exception as e:
+            print(f"Error getting bbox for {full_name}: {e}")
+
+    def bbox_to_img_coords(bbox):
+        """Convert matplotlib bbox to image pixel coordinates."""
+        x0_inches = bbox.x0 / fig.dpi
+        y0_inches = bbox.y0 / fig.dpi
+        x1_inches = bbox.x1 / fig.dpi
+        y1_inches = bbox.y1 / fig.dpi
+        x0_rel = x0_inches - tight_x0 + pad_inches
+        y0_rel = y0_inches - tight_y0 + pad_inches
+        x1_rel = x1_inches - tight_x0 + pad_inches
+        y1_rel = y1_inches - tight_y0 + pad_inches
+        return {
+            "x0": int(x0_rel * scale_x),
+            "y0": int((saved_height_inches - y1_rel) * scale_y),
+            "x1": int(x1_rel * scale_x),
+            "y1": int((saved_height_inches - y0_rel) * scale_y),
+        }
+
+    # Extract bboxes for each axis
+    for ax_id, ax in axes_map.items():
+        # Get axes bounding box (the entire panel area)
+        try:
+            ax_bbox = ax.get_window_extent(renderer)
+            coords = bbox_to_img_coords(ax_bbox)
+            # Extract actual title/labels from the axes
+            title_text = ax.title.get_text() if ax.title else ""
+            xlabel_text = ax.xaxis.label.get_text() if ax.xaxis.label else ""
+            ylabel_text = ax.yaxis.label.get_text() if ax.yaxis.label else ""
+            bboxes[f"{ax_id}_panel"] = {
+                **coords,
+                "label": f"Panel {ax_id}",
+                "ax_id": ax_id,
+                "is_panel": True,
+                "title": title_text,
+                "xlabel": xlabel_text,
+                "ylabel": ylabel_text,
+            }
+        except Exception as e:
+            print(f"Error getting panel bbox for {ax_id}: {e}")
+
+        # Get bboxes for title, labels
+        if ax.title.get_text():
+            get_element_bbox(ax.title, "title", ax_id, ax)
+        if ax.xaxis.label.get_text():
+            get_element_bbox(ax.xaxis.label, "xlabel", ax_id, ax)
+        if ax.yaxis.label.get_text():
+            get_element_bbox(ax.yaxis.label, "ylabel", ax_id, ax)
+
+        # Get X-axis bbox (spine + ticks + ticklabels)
+        _extract_axis_bboxes_for_axis(
+            ax, ax_id, renderer, bboxes, bbox_to_img_coords, Bbox
+        )
+
+        # Get legend bbox
+        legend = ax.get_legend()
+        if legend:
+            get_element_bbox(legend, "legend", ax_id, ax)
+            # Add element_type for drag detection
+            if f"{ax_id}_legend" in bboxes:
+                bboxes[f"{ax_id}_legend"]["element_type"] = "legend"
+                bboxes[f"{ax_id}_legend"]["draggable"] = True
+
+        # Get panel letter (text annotations like A, B, C)
+        import re
+
+        panel_letter_pattern = re.compile(r"^[A-Z]\.?$|^\([A-Za-z]\)$")
+        for idx, text_artist in enumerate(ax.texts):
+            text_content = text_artist.get_text().strip()
+            if text_content and panel_letter_pattern.match(text_content):
+                name = f"panel_letter_{text_content.replace('.', '').replace('(', '').replace(')', '')}"
+                get_element_bbox(text_artist, name, ax_id, ax)
+                full_name = f"{ax_id}_{name}"
+                if full_name in bboxes:
+                    bboxes[full_name]["element_type"] = "panel_letter"
+                    bboxes[full_name]["draggable"] = True
+                    bboxes[full_name]["text"] = text_content
+                    # Get position in axes coordinates (0-1)
+                    pos = text_artist.get_position()
+                    transform = text_artist.get_transform()
+                    if transform == ax.transAxes:
+                        bboxes[full_name]["axes_position"] = {"x": pos[0], "y": pos[1]}
+
+        # Get trace (line) bboxes
+        _extract_trace_bboxes_for_axis(
+            ax,
+            ax_id,
+            fig,
+            renderer,
+            bboxes,
+            get_element_bbox,
+            tight_x0,
+            tight_y0,
+            saved_height_inches,
+            scale_x,
+            scale_y,
+            pad_inches,
+        )
+
+    # Get caption bbox (figure-level text)
+    # This is outside the per-axis loop since caption is a figure-level element
+    for text_artist in fig.texts:
+        text_content = text_artist.get_text()
+        if text_content:
+            pos = text_artist.get_position()
+            # Caption is typically centered horizontally (x ~= 0.5) and below figure (y < 0.1)
+            if pos[1] < 0.1:
+                try:
+                    bbox = text_artist.get_window_extent(renderer)
+                    x0_inches = bbox.x0 / fig.dpi
+                    x1_inches = bbox.x1 / fig.dpi
+                    y0_inches = bbox.y0 / fig.dpi
+                    y1_inches = bbox.y1 / fig.dpi
+                    x0_rel = x0_inches - tight_x0 + pad_inches
+                    x1_rel = x1_inches - tight_x0 + pad_inches
+                    y0_rel = saved_height_inches - (y1_inches - tight_y0 + pad_inches)
+                    y1_rel = saved_height_inches - (y0_inches - tight_y0 + pad_inches)
+                    bboxes["caption"] = {
+                        "x0": max(0, int(x0_rel * scale_x)),
+                        "y0": max(0, int(y0_rel * scale_y)),
+                        "x1": min(img_width, int(x1_rel * scale_x)),
+                        "y1": min(img_height, int(y1_rel * scale_y)),
+                        "label": "Caption",
+                    }
+                except Exception as e:
+                    print(f"Error getting caption bbox: {e}")
+                break  # Only one caption expected
+
+    # Add schema v0.3 metadata if available
+    if GEOMETRY_V03_AVAILABLE:
+        axes_bboxes = {}
+        for ax_id, ax in axes_map.items():
+            axes_bboxes[ax_id] = extract_axes_bbox_px(ax, fig)
+        bboxes["_meta"] = {
+            "schema_version": "0.3.0",
+            "axes": axes_bboxes,
+            "geometry_available": True,
+        }
+
+    return bboxes
+
+
+def _extract_trace_bboxes_for_axis(
+    ax,
+    ax_id,
+    fig,
+    renderer,
+    bboxes,
+    get_element_bbox,
+    tight_x0,
+    tight_y0,
+    saved_height_inches,
+    scale_x,
+    scale_y,
+    pad_inches,
+):
+    """Extract bboxes for all data elements in a specific axis.
+
+    Handles:
+    - Lines (plot, errorbar lines)
+    - Scatter points (PathCollection)
+    - Fill areas (PolyCollection from fill_between)
+    - Bars (Rectangle patches)
+    """
+    import numpy as np
+
+    def coords_to_img_points(data_coords):
+        """Convert data coordinates to image pixel coordinates."""
+        if len(data_coords) == 0:
+            return []
+        transform = ax.transData
+        points_display = transform.transform(data_coords)
+        points_img = []
+        for px, py in points_display:
+            # Skip invalid points (NaN, infinity)
+            if not np.isfinite(px) or not np.isfinite(py):
+                continue
+            px_inches = px / fig.dpi
+            py_inches = py / fig.dpi
+            x_rel = px_inches - tight_x0 + pad_inches
+            y_rel = saved_height_inches - (py_inches - tight_y0 + pad_inches)
+            # Clamp to reasonable bounds to avoid overflow
+            x_img = max(-10000, min(10000, int(x_rel * scale_x)))
+            y_img = max(-10000, min(10000, int(y_rel * scale_y)))
+            points_img.append([x_img, y_img])
+        # Downsample if too many
+        if len(points_img) > 100:
+            step = len(points_img) // 100
+            points_img = points_img[::step]
+        return points_img
+
+    # 1. Extract lines (plot, errorbar lines, etc.)
+    line_idx = 0
+    for line in ax.get_lines():
+        try:
+            label = line.get_label()
+            # Include unlabeled lines but mark them appropriately
+            if label is None or label.startswith("_"):
+                label = None  # Will use generic name
+
+            trace_name = f"trace_{line_idx}"
+            full_name = f"{ax_id}_{trace_name}"
+            get_element_bbox(line, trace_name, ax_id, ax)
+
+            if full_name in bboxes:
+                bboxes[full_name]["label"] = f"{ax_id}: {label or f'Line {line_idx}'}"
+                bboxes[full_name]["trace_idx"] = line_idx
+                bboxes[full_name]["element_type"] = "line"
+
+                xdata, ydata = line.get_xdata(), line.get_ydata()
+                if len(xdata) > 0:
+                    bboxes[full_name]["points"] = coords_to_img_points(
+                        list(zip(xdata, ydata))
+                    )
+
+                # Add schema v0.3 geometry_px if available
+                if GEOMETRY_V03_AVAILABLE:
+                    try:
+                        geom = extract_line_geometry(line, ax, fig)
+                        bboxes[full_name]["geometry_px"] = geom
+                    except Exception:
+                        pass  # Fall back to legacy points
+
+            line_idx += 1
+        except Exception as e:
+            print(f"Error getting line bbox for {ax_id}: {e}")
+
+    # 2. Extract collections (scatter, fill_between, etc.)
+    coll_idx = 0
+    for coll in ax.collections:
+        try:
+            label = coll.get_label()
+            if label is None or label.startswith("_"):
+                # Still extract unlabeled collections but with generic name
+                label = None
+
+            coll_type = type(coll).__name__
+            if coll_type == "PathCollection":
+                # Scatter points
+                element_name = f"scatter_{coll_idx}"
+                full_name = f"{ax_id}_{element_name}"
+                get_element_bbox(coll, element_name, ax_id, ax)
+
+                if full_name in bboxes:
+                    bboxes[full_name][
+                        "label"
+                    ] = f"{ax_id}: {label or f'Scatter {coll_idx}'}"
+                    bboxes[full_name]["element_type"] = "scatter"
+
+                    # Get scatter point positions
+                    offsets = coll.get_offsets()
+                    if len(offsets) > 0:
+                        bboxes[full_name]["points"] = coords_to_img_points(offsets)
+
+                    # Add schema v0.3 geometry_px if available
+                    if GEOMETRY_V03_AVAILABLE:
+                        try:
+                            geom = extract_scatter_geometry(coll, ax, fig)
+                            bboxes[full_name]["geometry_px"] = geom
+                        except Exception:
+                            pass  # Fall back to legacy points
+
+            elif coll_type in ("PolyCollection", "FillBetweenPolyCollection"):
+                # Fill areas (fill_between, etc.)
+                element_name = f"fill_{coll_idx}"
+                full_name = f"{ax_id}_{element_name}"
+                get_element_bbox(coll, element_name, ax_id, ax)
+
+                if full_name in bboxes:
+                    bboxes[full_name][
+                        "label"
+                    ] = f"{ax_id}: {label or f'Fill {coll_idx}'}"
+                    bboxes[full_name]["element_type"] = "fill"
+
+                    # Add schema v0.3 geometry_px if available
+                    if GEOMETRY_V03_AVAILABLE:
+                        try:
+                            geom = extract_polygon_geometry(coll, ax, fig)
+                            bboxes[full_name]["geometry_px"] = geom
+                        except Exception:
+                            pass
+
+            coll_idx += 1
+        except Exception as e:
+            print(f"Error getting collection bbox for {ax_id}: {e}")
+
+    # 3. Extract patches (bars, rectangles, etc.)
+    patch_idx = 0
+    for patch in ax.patches:
+        try:
+            label = patch.get_label()
+            patch_type = type(patch).__name__
+
+            if patch_type == "Rectangle":
+                # Bar chart bars
+                element_name = f"bar_{patch_idx}"
+                full_name = f"{ax_id}_{element_name}"
+                get_element_bbox(patch, element_name, ax_id, ax)
+
+                if full_name in bboxes:
+                    bboxes[full_name][
+                        "label"
+                    ] = f"{ax_id}: {label or f'Bar {patch_idx}'}"
+                    bboxes[full_name]["element_type"] = "bar"
+
+            patch_idx += 1
+        except Exception as e:
+            print(f"Error getting patch bbox for {ax_id}: {e}")
+
+
+def _extract_axis_bboxes_for_axis(
+    ax, ax_id, renderer, bboxes, bbox_to_img_coords, Bbox
+):
+    """Extract X and Y axis bboxes for a specific axis (multi-axis version)."""
+    try:
+        # X-axis: combine spine and tick labels into one bbox
+        x_axis_bboxes = []
+        for ticklabel in ax.xaxis.get_ticklabels():
+            if ticklabel.get_visible():
+                try:
+                    tb = ticklabel.get_window_extent(renderer)
+                    if tb.width > 0:
+                        x_axis_bboxes.append(tb)
+                except Exception:
+                    pass
+        for tick in ax.xaxis.get_major_ticks():
+            if tick.tick1line.get_visible():
+                try:
+                    tb = tick.tick1line.get_window_extent(renderer)
+                    if tb.width > 0 or tb.height > 0:
+                        x_axis_bboxes.append(tb)
+                except Exception:
+                    pass
+        spine_bbox = ax.spines["bottom"].get_window_extent(renderer)
+        if spine_bbox.width > 0:
+            if x_axis_bboxes:
+                tick_union = Bbox.union(x_axis_bboxes)
+                constrained_spine = Bbox.from_extents(
+                    tick_union.x0, spine_bbox.y0, tick_union.x1, spine_bbox.y1
+                )
+                x_axis_bboxes.append(constrained_spine)
+            else:
+                x_axis_bboxes.append(spine_bbox)
+        if x_axis_bboxes:
+            combined = Bbox.union(x_axis_bboxes)
+            bboxes[f"{ax_id}_xaxis"] = bbox_to_img_coords(combined)
+            bboxes[f"{ax_id}_xaxis"]["label"] = f"{ax_id}: X Axis"
+            bboxes[f"{ax_id}_xaxis"]["ax_id"] = ax_id
+            bboxes[f"{ax_id}_xaxis"]["element_type"] = "xaxis"
+
+        # Y-axis: combine spine and tick labels into one bbox
+        y_axis_bboxes = []
+        for ticklabel in ax.yaxis.get_ticklabels():
+            if ticklabel.get_visible():
+                try:
+                    tb = ticklabel.get_window_extent(renderer)
+                    if tb.width > 0:
+                        y_axis_bboxes.append(tb)
+                except Exception:
+                    pass
+        for tick in ax.yaxis.get_major_ticks():
+            if tick.tick1line.get_visible():
+                try:
+                    tb = tick.tick1line.get_window_extent(renderer)
+                    if tb.width > 0 or tb.height > 0:
+                        y_axis_bboxes.append(tb)
+                except Exception:
+                    pass
+        spine_bbox = ax.spines["left"].get_window_extent(renderer)
+        if spine_bbox.height > 0:
+            if y_axis_bboxes:
+                tick_union = Bbox.union(y_axis_bboxes)
+                constrained_spine = Bbox.from_extents(
+                    spine_bbox.x0, tick_union.y0, spine_bbox.x1, tick_union.y1
+                )
+                y_axis_bboxes.append(constrained_spine)
+            else:
+                y_axis_bboxes.append(spine_bbox)
+        if y_axis_bboxes:
+            combined = Bbox.union(y_axis_bboxes)
+            padded = Bbox.from_extents(
+                combined.x0 - 10, combined.y0 - 5, combined.x1 + 5, combined.y1 + 5
+            )
+            bboxes[f"{ax_id}_yaxis"] = bbox_to_img_coords(padded)
+            bboxes[f"{ax_id}_yaxis"]["label"] = f"{ax_id}: Y Axis"
+            bboxes[f"{ax_id}_yaxis"]["ax_id"] = ax_id
+            bboxes[f"{ax_id}_yaxis"]["element_type"] = "yaxis"
+
+    except Exception as e:
+        print(f"Error getting axis bboxes for {ax_id}: {e}")
+
+
+def _extract_axis_bboxes(ax, renderer, bboxes, bbox_to_img_coords, Bbox, ax_prefix=""):
+    """Extract bboxes for X and Y axis elements.
+
+    Args:
+        ax: Matplotlib axis.
+        renderer: Figure renderer.
+        bboxes: Dict to store bboxes.
+        bbox_to_img_coords: Coordinate conversion function.
+        Bbox: Matplotlib Bbox class.
+        ax_prefix: Prefix for bbox names (e.g., "ax_00_").
+    """
+    try:
+        # X-axis: combine spine and tick labels into one bbox
+        x_axis_bboxes = []
+        for ticklabel in ax.xaxis.get_ticklabels():
+            if ticklabel.get_visible():
+                try:
+                    tb = ticklabel.get_window_extent(renderer)
+                    if tb.width > 0:
+                        x_axis_bboxes.append(tb)
+                except Exception:
+                    pass
+        for tick in ax.xaxis.get_major_ticks():
+            if tick.tick1line.get_visible():
+                try:
+                    tb = tick.tick1line.get_window_extent(renderer)
+                    if tb.width > 0 or tb.height > 0:
+                        x_axis_bboxes.append(tb)
+                except Exception:
+                    pass
+        spine_bbox = ax.spines["bottom"].get_window_extent(renderer)
+        if spine_bbox.width > 0:
+            if x_axis_bboxes:
+                tick_union = Bbox.union(x_axis_bboxes)
+                constrained_spine = Bbox.from_extents(
+                    tick_union.x0, spine_bbox.y0, tick_union.x1, spine_bbox.y1
+                )
+                x_axis_bboxes.append(constrained_spine)
+            else:
+                x_axis_bboxes.append(spine_bbox)
+        if x_axis_bboxes:
+            combined = Bbox.union(x_axis_bboxes)
+            bboxes[f"{ax_prefix}xaxis_spine"] = bbox_to_img_coords(combined)
+            bboxes[f"{ax_prefix}xaxis_spine"]["label"] = "X Spine & Ticks"
+
+        # Y-axis: combine spine and tick labels into one bbox
+        y_axis_bboxes = []
+        for ticklabel in ax.yaxis.get_ticklabels():
+            if ticklabel.get_visible():
+                try:
+                    tb = ticklabel.get_window_extent(renderer)
+                    if tb.width > 0:
+                        y_axis_bboxes.append(tb)
+                except Exception:
+                    pass
+        for tick in ax.yaxis.get_major_ticks():
+            if tick.tick1line.get_visible():
+                try:
+                    tb = tick.tick1line.get_window_extent(renderer)
+                    if tb.width > 0 or tb.height > 0:
+                        y_axis_bboxes.append(tb)
+                except Exception:
+                    pass
+        spine_bbox = ax.spines["left"].get_window_extent(renderer)
+        if spine_bbox.height > 0:
+            if y_axis_bboxes:
+                tick_union = Bbox.union(y_axis_bboxes)
+                constrained_spine = Bbox.from_extents(
+                    spine_bbox.x0, tick_union.y0, spine_bbox.x1, tick_union.y1
+                )
+                y_axis_bboxes.append(constrained_spine)
+            else:
+                y_axis_bboxes.append(spine_bbox)
+        if y_axis_bboxes:
+            combined = Bbox.union(y_axis_bboxes)
+            padded = Bbox.from_extents(
+                combined.x0 - 10, combined.y0 - 5, combined.x1 + 5, combined.y1 + 5
+            )
+            bboxes[f"{ax_prefix}yaxis_spine"] = bbox_to_img_coords(padded)
+            bboxes[f"{ax_prefix}yaxis_spine"]["label"] = "Y Spine & Ticks"
+
+    except Exception as e:
+        print(f"Error getting axis bboxes: {e}")
+
+
+def _extract_trace_bboxes(
+    ax,
+    fig,
+    renderer,
+    bboxes,
+    get_element_bbox,
+    tight_x0,
+    tight_y0,
+    saved_height_inches,
+    scale_x,
+    scale_y,
+    pad_inches,
+):
+    """Extract bboxes for all data elements (lines, scatter, fill) with proximity detection."""
+    import numpy as np
+
+    def coords_to_img_points(data_coords):
+        """Convert data coordinates to image pixel coordinates."""
+        if len(data_coords) == 0:
+            return []
+        transform = ax.transData
+        points_display = transform.transform(data_coords)
+        points_img = []
+        for px, py in points_display:
+            if not np.isfinite(px) or not np.isfinite(py):
+                continue
+            px_inches = px / fig.dpi
+            py_inches = py / fig.dpi
+            x_rel = px_inches - tight_x0 + pad_inches
+            y_rel = saved_height_inches - (py_inches - tight_y0 + pad_inches)
+            x_img = max(-10000, min(10000, int(x_rel * scale_x)))
+            y_img = max(-10000, min(10000, int(y_rel * scale_y)))
+            points_img.append([x_img, y_img])
+        if len(points_img) > 100:
+            step = len(points_img) // 100
+            points_img = points_img[::step]
+        return points_img
+
+    # 1. Extract lines
+    for idx, line in enumerate(ax.get_lines()):
+        try:
+            label = line.get_label()
+            # Include unlabeled lines but mark them appropriately
+            if label is None or label.startswith("_"):
+                label = None  # Will use generic name
+            get_element_bbox(line, f"trace_{idx}")
+            if f"trace_{idx}" in bboxes:
+                bboxes[f"trace_{idx}"]["label"] = label or f"Trace {idx}"
+                bboxes[f"trace_{idx}"]["trace_idx"] = idx
+                bboxes[f"trace_{idx}"]["element_type"] = "line"
+
+                xdata, ydata = line.get_xdata(), line.get_ydata()
+                if len(xdata) > 0:
+                    bboxes[f"trace_{idx}"]["points"] = coords_to_img_points(
+                        list(zip(xdata, ydata))
+                    )
+
+                # Add schema v0.3 geometry_px if available
+                if GEOMETRY_V03_AVAILABLE:
+                    try:
+                        geom = extract_line_geometry(line, ax, fig)
+                        bboxes[f"trace_{idx}"]["geometry_px"] = geom
+                    except Exception:
+                        pass
+        except Exception as e:
+            print(f"Error getting trace bbox: {e}")
+
+    # 2. Extract collections (scatter, fill_between)
+    coll_idx = 0
+    for coll in ax.collections:
+        try:
+            label = coll.get_label()
+            if label is None or label.startswith("_"):
+                label = None
+
+            coll_type = type(coll).__name__
+            if coll_type == "PathCollection":
+                # Scatter points
+                elem_key = f"scatter_{coll_idx}"
+                get_element_bbox(coll, elem_key)
+
+                # Initialize entry if bbox extraction failed but we have data
+                offsets = coll.get_offsets()
+                if elem_key not in bboxes and len(offsets) > 0:
+                    # Create bbox from data coordinates as fallback
+                    points_img = coords_to_img_points(offsets)
+                    if points_img:
+                        xs = [p[0] for p in points_img]
+                        ys = [p[1] for p in points_img]
+                        bboxes[elem_key] = {
+                            "x0": min(xs) - 10,
+                            "y0": min(ys) - 10,
+                            "x1": max(xs) + 10,
+                            "y1": max(ys) + 10,
+                        }
+
+                if elem_key in bboxes:
+                    bboxes[elem_key]["label"] = label or f"Scatter {coll_idx}"
+                    bboxes[elem_key]["element_type"] = "scatter"
+
+                    if len(offsets) > 0:
+                        bboxes[elem_key]["points"] = coords_to_img_points(offsets)
+
+                    # Add schema v0.3 geometry_px if available
+                    if GEOMETRY_V03_AVAILABLE:
+                        try:
+                            geom = extract_scatter_geometry(coll, ax, fig)
+                            bboxes[elem_key]["geometry_px"] = geom
+                        except Exception:
+                            pass
+
+            elif coll_type in ("PolyCollection", "FillBetweenPolyCollection"):
+                # Fill areas
+                get_element_bbox(coll, f"fill_{coll_idx}")
+                if f"fill_{coll_idx}" in bboxes:
+                    bboxes[f"fill_{coll_idx}"]["label"] = label or f"Fill {coll_idx}"
+                    bboxes[f"fill_{coll_idx}"]["element_type"] = "fill"
+
+                    # Add schema v0.3 geometry_px if available
+                    if GEOMETRY_V03_AVAILABLE:
+                        try:
+                            geom = extract_polygon_geometry(coll, ax, fig)
+                            bboxes[f"fill_{coll_idx}"]["geometry_px"] = geom
+                        except Exception:
+                            pass
+
+            coll_idx += 1
+        except Exception as e:
+            print(f"Error getting collection bbox: {e}")
+
+
+def extract_bboxes_from_metadata(
+    metadata: Dict[str, Any],
+    img_width: int,
+    img_height: int,
+) -> Dict[str, Any]:
+    """Extract bounding boxes from pre-computed metadata (without re-rendering).
+
+    This is used when loading actual PNGs from bundles instead of re-rendering.
+    Extracts bbox info from:
+    - hit_regions (if available from v0.3 schema)
+    - elements dict
+    - axes positions
+
+    Args:
+        metadata: JSON metadata from spec.json or panel JSON
+        img_width: Image width in pixels
+        img_height: Image height in pixels
+
+    Returns:
+        Dict with bboxes keyed by element name
+    """
+    bboxes = {}
+
+    # Check for pre-computed hit_regions (v0.3 schema)
+    hit_regions = metadata.get("hit_regions", {})
+    if hit_regions:
+        color_map = hit_regions.get("color_map", {})
+        for element_name, color in color_map.items():
+            # We don't have exact coords from color map, but we can create placeholder
+            bboxes[element_name] = {
+                "label": element_name.replace("_", " ").title(),
+                "element_type": _guess_element_type(element_name),
+            }
+
+    # Check for geometry_px in cache (v0.3 layered bundle)
+    geometry_px = metadata.get("geometry_px", {})
+    if geometry_px:
+        for element_name, geom in geometry_px.items():
+            if isinstance(geom, dict) and "bbox" in geom:
+                bbox = geom["bbox"]
+                bboxes[element_name] = {
+                    "x0": bbox.get("x0", 0),
+                    "y0": bbox.get("y0", 0),
+                    "x1": bbox.get("x1", img_width),
+                    "y1": bbox.get("y1", img_height),
+                    "label": element_name.replace("_", " ").title(),
+                    "element_type": _guess_element_type(element_name),
+                }
+                if "points" in geom:
+                    bboxes[element_name]["points"] = geom["points"]
+
+    # Extract from elements dict if present
+    elements = metadata.get("elements", {})
+    if not isinstance(elements, dict):
+        elements = {}
+    for element_name, element_info in elements.items():
+        if not isinstance(element_info, dict):
+            continue
+        if element_name not in bboxes:
+            bboxes[element_name] = {
+                "label": element_info.get(
+                    "label", element_name.replace("_", " ").title()
+                ),
+                "element_type": element_info.get(
+                    "type", _guess_element_type(element_name)
+                ),
+            }
+
+    # Extract from axes (handle both dict and list formats)
+    axes = metadata.get("axes", [])
+    if isinstance(axes, list):
+        axes_list = axes
+    elif isinstance(axes, dict):
+        axes_list = list(axes.values())
+    else:
+        axes_list = []
+
+    for i, ax_spec in enumerate(axes_list):
+        if not isinstance(ax_spec, dict):
+            continue
+
+        ax_id = ax_spec.get("id", f"ax{i}")
+
+        # Panel bbox - check for "bbox" field (new format) or "position" (old format)
+        bbox_spec = ax_spec.get("bbox", {})
+        pos = ax_spec.get("position", [])
+
+        if bbox_spec and isinstance(bbox_spec, dict):
+            # New format: bbox with x0, y0, width, height in panel fraction
+            x0_frac = bbox_spec.get("x0", 0)
+            y0_frac = bbox_spec.get("y0", 0)
+            w_frac = bbox_spec.get("width", 1)
+            h_frac = bbox_spec.get("height", 1)
+            x0 = int(x0_frac * img_width)
+            y0 = int(y0_frac * img_height)
+            x1 = int((x0_frac + w_frac) * img_width)
+            y1 = int((y0_frac + h_frac) * img_height)
+            bboxes[f"{ax_id}_panel"] = {
+                "x0": x0,
+                "y0": y0,
+                "x1": x1,
+                "y1": y1,
+                "label": f"Panel {ax_id}",
+                "ax_id": ax_id,
+                "is_panel": True,
+            }
+        elif len(pos) >= 4:
+            # Old format: position is in figure fraction [x0, y0, width, height]
+            x0 = int(pos[0] * img_width)
+            y0 = int((1 - pos[1] - pos[3]) * img_height)  # Flip Y
+            x1 = int((pos[0] + pos[2]) * img_width)
+            y1 = int((1 - pos[1]) * img_height)  # Flip Y
+            bboxes[f"{ax_id}_panel"] = {
+                "x0": x0,
+                "y0": y0,
+                "x1": x1,
+                "y1": y1,
+                "label": f"Panel {ax_id}",
+                "ax_id": ax_id,
+                "is_panel": True,
+            }
+
+        # Title/labels from labels dict (new format) or xaxis/yaxis (old format)
+        labels = ax_spec.get("labels", {})
+        xaxis = ax_spec.get("xaxis", {})
+        yaxis = ax_spec.get("yaxis", {})
+
+        xlabel = labels.get("xlabel") or (
+            xaxis.get("label") if isinstance(xaxis, dict) else None
+        )
+        ylabel = labels.get("ylabel") or (
+            yaxis.get("label") if isinstance(yaxis, dict) else None
+        )
+        title = labels.get("title")
+
+        if xlabel:
+            bboxes[f"{ax_id}_xlabel"] = {
+                "label": f"{ax_id}: {xlabel}",
+                "element_type": "xlabel",
+                "ax_id": ax_id,
+            }
+        if ylabel:
+            bboxes[f"{ax_id}_ylabel"] = {
+                "label": f"{ax_id}: {ylabel}",
+                "element_type": "ylabel",
+                "ax_id": ax_id,
+            }
+        if title:
+            bboxes[f"{ax_id}_title"] = {
+                "label": title,
+                "element_type": "title",
+                "ax_id": ax_id,
+            }
+
+    # Extract from traces array (pltz spec format)
+    traces = metadata.get("traces", [])
+    if isinstance(traces, list):
+        for i, trace in enumerate(traces):
+            if not isinstance(trace, dict):
+                continue
+            trace_id = trace.get("id", f"trace_{i}")
+            trace_type = trace.get("type", "line")
+            trace_label = trace.get("label", f"Trace {i}")
+            ax_idx = trace.get("axes_index", 0)
+
+            # Use axes bbox as fallback for trace bbox
+            ax_panel_key = None
+            for key in bboxes:
+                if key.endswith("_panel") and bboxes[key].get("ax_id", "").endswith(
+                    str(ax_idx)
+                ):
+                    ax_panel_key = key
+                    break
+            if not ax_panel_key:
+                # Find any panel bbox
+                for key in bboxes:
+                    if key.endswith("_panel"):
+                        ax_panel_key = key
+                        break
+
+            trace_bbox = {
+                "label": trace_label,
+                "element_type": trace_type,
+                "trace_idx": i,
+                "ax_id": f"ax{ax_idx}",
+            }
+
+            # Copy panel bbox coordinates if available
+            if ax_panel_key and ax_panel_key in bboxes:
+                panel = bboxes[ax_panel_key]
+                trace_bbox["x0"] = panel.get("x0", 0)
+                trace_bbox["y0"] = panel.get("y0", 0)
+                trace_bbox["x1"] = panel.get("x1", img_width)
+                trace_bbox["y1"] = panel.get("y1", img_height)
+
+            bboxes[f"trace_{i}"] = trace_bbox
+
+    # If no bboxes found, return minimal set
+    if not bboxes:
+        bboxes["panel"] = {
+            "x0": 0,
+            "y0": 0,
+            "x1": img_width,
+            "y1": img_height,
+            "label": "Panel",
+            "is_panel": True,
+        }
+
+    return bboxes
+
+
+def _guess_element_type(name: str) -> str:
+    """Guess element type from element name."""
+    name_lower = name.lower()
+    if "line" in name_lower or "trace" in name_lower:
+        return "line"
+    elif "scatter" in name_lower:
+        return "scatter"
+    elif "bar" in name_lower:
+        return "bar"
+    elif "fill" in name_lower:
+        return "fill"
+    elif "xlabel" in name_lower:
+        return "xlabel"
+    elif "ylabel" in name_lower:
+        return "ylabel"
+    elif "title" in name_lower:
+        return "title"
+    elif "legend" in name_lower:
+        return "legend"
+    elif "xaxis" in name_lower:
+        return "xaxis"
+    elif "yaxis" in name_lower:
+        return "yaxis"
+    elif "panel" in name_lower:
+        return "panel"
+    return "unknown"
+
+
+def extract_bboxes_from_geometry_px(
+    geometry_data: Dict[str, Any],
+    img_width: int,
+    img_height: int,
+) -> Dict[str, Any]:
+    """Extract bounding boxes from geometry_px.json (cached pixel coordinates).
+
+    This provides precise pixel coordinates for interactive element selection.
+
+    Args:
+        geometry_data: JSON data from geometry_px.json
+        img_width: Actual image width in pixels
+        img_height: Actual image height in pixels
+
+    Returns:
+        Dict with bboxes keyed by element name
+    """
+    bboxes = {}
+
+    # Get figure dimensions from geometry to calculate scale
+    figure_px = geometry_data.get("figure_px", [img_width, img_height])
+    if isinstance(figure_px, list) and len(figure_px) >= 2:
+        geom_width, geom_height = figure_px[0], figure_px[1]
+    else:
+        geom_width, geom_height = img_width, img_height
+
+    # Scale factor if image size differs from geometry
+    scale_x = img_width / geom_width if geom_width > 0 else 1
+    scale_y = img_height / geom_height if geom_height > 0 else 1
+
+    # Extract axes bboxes
+    axes = geometry_data.get("axes", [])
+    for i, ax in enumerate(axes):
+        if not isinstance(ax, dict):
+            continue
+        ax_id = ax.get("id", f"ax{i}")
+        bbox_px = ax.get("bbox_px", {})
+        if bbox_px:
+            x0 = float(bbox_px.get("x0", 0)) * scale_x
+            y0 = float(bbox_px.get("y0", 0)) * scale_y
+            w = float(bbox_px.get("width", 0)) * scale_x
+            h = float(bbox_px.get("height", 0)) * scale_y
+            bboxes[f"{ax_id}_panel"] = {
+                "x0": int(x0),
+                "y0": int(y0),
+                "x1": int(x0 + w),
+                "y1": int(y0 + h),
+                "label": f"Axes {ax_id}",
+                "ax_id": ax_id,
+                "is_panel": True,
+            }
+
+    # Helper to safely convert to int (handle inf/nan)
+    def safe_int(val, default=0, max_val=10000):
+        import math
+
+        if val is None or math.isinf(val) or math.isnan(val):
+            return default
+        return max(0, min(int(val), max_val))
+
+    # Extract artists (lines, scatter, bars, etc.)
+    artists = geometry_data.get("artists", [])
+    for i, artist in enumerate(artists):
+        if not isinstance(artist, dict):
+            continue
+
+        artist_id = artist.get("id", str(i))
+        artist_type = artist.get("type", "unknown")
+        artist_label = artist.get("label") or f"{artist_type}_{i}"
+        axes_index = artist.get("axes_index", 0)
+
+        # Get bbox_px
+        bbox_px = artist.get("bbox_px", {})
+        if bbox_px:
+            x0 = float(bbox_px.get("x0", 0)) * scale_x
+            y0 = float(bbox_px.get("y0", 0)) * scale_y
+            w = float(bbox_px.get("width", 0)) * scale_x
+            h = float(bbox_px.get("height", 0)) * scale_y
+
+            artist_bbox = {
+                "x0": safe_int(x0, 0, img_width),
+                "y0": safe_int(y0, 0, img_height),
+                "x1": safe_int(x0 + w, img_width, img_width),
+                "y1": safe_int(y0 + h, img_height, img_height),
+                "label": artist_label,
+                "element_type": artist_type,
+                "trace_idx": i,
+                "ax_id": f"ax{axes_index}",
+            }
+
+            # Get path_px for lines (for precise hover detection)
+            path_px = artist.get("path_px", [])
+            if path_px and len(path_px) > 0:
+                import math
+
+                # Scale points to actual image coordinates, filter out inf/nan
+                scaled_points = []
+                for pt in path_px:
+                    if isinstance(pt, (list, tuple)) and len(pt) >= 2:
+                        px, py = pt[0] * scale_x, pt[1] * scale_y
+                        if not (
+                            math.isinf(px)
+                            or math.isinf(py)
+                            or math.isnan(px)
+                            or math.isnan(py)
+                        ):
+                            scaled_points.append([px, py])
+                if scaled_points:
+                    artist_bbox["points"] = scaled_points
+
+            # Get scatter points if available
+            scatter_px = artist.get("scatter_px", [])
+            if scatter_px and len(scatter_px) > 0:
+                import math
+
+                scaled_points = []
+                for pt in scatter_px:
+                    if isinstance(pt, (list, tuple)) and len(pt) >= 2:
+                        px, py = pt[0] * scale_x, pt[1] * scale_y
+                        if not (
+                            math.isinf(px)
+                            or math.isinf(py)
+                            or math.isnan(px)
+                            or math.isnan(py)
+                        ):
+                            scaled_points.append([px, py])
+                if scaled_points:
+                    artist_bbox["points"] = scaled_points
+                    artist_bbox["element_type"] = "scatter"
+
+            bboxes[f"trace_{i}"] = artist_bbox
+
+    # Extract from selectable_regions (title, xlabel, ylabel, xaxis, yaxis)
+    selectable = geometry_data.get("selectable_regions", {})
+    sel_axes = selectable.get("axes", [])
+    for ax_data in sel_axes:
+        if not isinstance(ax_data, dict):
+            continue
+        ax_idx = ax_data.get("index", 0)
+        ax_id = f"ax{ax_idx}"
+
+        # Title
+        title_data = ax_data.get("title", {})
+        if title_data and "bbox_px" in title_data:
+            bbox = title_data["bbox_px"]
+            if isinstance(bbox, list) and len(bbox) >= 4:
+                bboxes[f"{ax_id}_title"] = {
+                    "x0": safe_int(bbox[0] * scale_x, 0, img_width),
+                    "y0": safe_int(bbox[1] * scale_y, 0, img_height),
+                    "x1": safe_int(bbox[2] * scale_x, img_width, img_width),
+                    "y1": safe_int(bbox[3] * scale_y, img_height, img_height),
+                    "label": title_data.get("text", "Title"),
+                    "element_type": "title",
+                    "ax_id": ax_id,
+                }
+
+        # X Label
+        xlabel_data = ax_data.get("xlabel", {})
+        if xlabel_data and "bbox_px" in xlabel_data:
+            bbox = xlabel_data["bbox_px"]
+            if isinstance(bbox, list) and len(bbox) >= 4:
+                bboxes[f"{ax_id}_xlabel"] = {
+                    "x0": safe_int(bbox[0] * scale_x, 0, img_width),
+                    "y0": safe_int(bbox[1] * scale_y, 0, img_height),
+                    "x1": safe_int(bbox[2] * scale_x, img_width, img_width),
+                    "y1": safe_int(bbox[3] * scale_y, img_height, img_height),
+                    "label": xlabel_data.get("text", "X Label"),
+                    "element_type": "xlabel",
+                    "ax_id": ax_id,
+                }
+
+        # Y Label
+        ylabel_data = ax_data.get("ylabel", {})
+        if ylabel_data and "bbox_px" in ylabel_data:
+            bbox = ylabel_data["bbox_px"]
+            if isinstance(bbox, list) and len(bbox) >= 4:
+                bboxes[f"{ax_id}_ylabel"] = {
+                    "x0": safe_int(bbox[0] * scale_x, 0, img_width),
+                    "y0": safe_int(bbox[1] * scale_y, 0, img_height),
+                    "x1": safe_int(bbox[2] * scale_x, img_width, img_width),
+                    "y1": safe_int(bbox[3] * scale_y, img_height, img_height),
+                    "label": ylabel_data.get("text", "Y Label"),
+                    "element_type": "ylabel",
+                    "ax_id": ax_id,
+                }
+
+        # X Axis spine
+        xaxis_data = ax_data.get("xaxis", {})
+        if xaxis_data:
+            spine = xaxis_data.get("spine", {})
+            if spine and "bbox_px" in spine:
+                bbox = spine["bbox_px"]
+                if isinstance(bbox, list) and len(bbox) >= 4:
+                    bboxes[f"{ax_id}_xaxis"] = {
+                        "x0": safe_int(bbox[0] * scale_x, 0, img_width),
+                        "y0": safe_int(bbox[1] * scale_y, 0, img_height),
+                        "x1": safe_int(bbox[2] * scale_x, img_width, img_width),
+                        "y1": safe_int(bbox[3] * scale_y, img_height, img_height),
+                        "label": "X Axis",
+                        "element_type": "xaxis",
+                        "ax_id": ax_id,
+                    }
+
+        # Y Axis spine
+        yaxis_data = ax_data.get("yaxis", {})
+        if yaxis_data:
+            spine = yaxis_data.get("spine", {})
+            if spine and "bbox_px" in spine:
+                bbox = spine["bbox_px"]
+                if isinstance(bbox, list) and len(bbox) >= 4:
+                    bboxes[f"{ax_id}_yaxis"] = {
+                        "x0": safe_int(bbox[0] * scale_x, 0, img_width),
+                        "y0": safe_int(bbox[1] * scale_y, 0, img_height),
+                        "x1": safe_int(bbox[2] * scale_x, img_width, img_width),
+                        "y1": safe_int(bbox[3] * scale_y, img_height, img_height),
+                        "label": "Y Axis",
+                        "element_type": "yaxis",
+                        "ax_id": ax_id,
+                    }
+
+        # Legend
+        legend_data = ax_data.get("legend", {})
+        if legend_data and "bbox_px" in legend_data:
+            bbox = legend_data["bbox_px"]
+            if isinstance(bbox, list) and len(bbox) >= 4:
+                bboxes[f"{ax_id}_legend"] = {
+                    "x0": safe_int(bbox[0] * scale_x, 0, img_width),
+                    "y0": safe_int(bbox[1] * scale_y, 0, img_height),
+                    "x1": safe_int(bbox[2] * scale_x, img_width, img_width),
+                    "y1": safe_int(bbox[3] * scale_y, img_height, img_height),
+                    "label": "Legend",
+                    "element_type": "legend",
+                    "ax_id": ax_id,
+                }
+
+    # If no bboxes found, return minimal set
+    if not bboxes:
+        bboxes["panel"] = {
+            "x0": 0,
+            "y0": 0,
+            "x1": img_width,
+            "y1": img_height,
+            "label": "Panel",
+            "is_panel": True,
+        }
+
+    return bboxes
+
+
+# EOF