fix(demo): stabilize visualizer bbox and mask rendering

Align bbox coordinate handling across primary and fallback paths, normalize Both-mode raw mask rendering, and tighten demo result typing to reduce runtime/display inconsistencies.

opengait/demo/window.py

@@ -12,10 +12,11 @@ import torch
 from jaxtyping import Float
 from numpy import ndarray
 
+from .preprocess import BBoxXYXY
+
 if TYPE_CHECKING:
     from numpy.typing import NDArray
 
-
 # Silhouette dimensions from preprocess.py
 SIL_HEIGHT: int = 64
 SIL_WIDTH: int = 44
@@ -239,19 +240,23 @@ def _to_numpy(obj: _ArrayLike) -> ndarray:
 
 def select_person(
     results: _DetectionResults,
-) -> tuple[ndarray, tuple[int, int, int, int], int] | None:
+) -> tuple[ndarray, BBoxXYXY, BBoxXYXY, int] | None:
     """Select the person with largest bounding box from detection results.
 
     Args:
         results: Detection results object with boxes and masks attributes.
             Expected to have:
-            - boxes.xyxy: array of bounding boxes [N, 4]
-            - masks.data: array of masks [N, H, W]
+            - boxes.xyxy: array of bounding boxes [N, 4] in frame coordinates (XYXY format)
+            - masks.data: array of masks [N, H, W] in mask coordinates
             - boxes.id: optional track IDs [N]
 
     Returns:
-        Tuple of (mask, bbox, track_id) for the largest person,
+        Tuple of (mask, bbox_mask, bbox_frame, track_id) for the largest person,
         or None if no valid detections or track IDs unavailable.
+        - mask: the person's segmentation mask
+        - bbox_mask: bounding box in mask coordinate space (XYXY format: x1, y1, x2, y2)
+        - bbox_frame: bounding box in frame coordinate space (XYXY format: x1, y1, x2, y2)
+        - track_id: the person's track ID
     """
     # Check for track IDs
     boxes_obj: _Boxes | object = getattr(results, "boxes", None)
@@ -329,20 +334,27 @@ def select_person(
         # Scale bbox from frame space to mask space
         scale_x = mask_w / frame_w if frame_w > 0 else 1.0
         scale_y = mask_h / frame_h if frame_h > 0 else 1.0
-        bbox = (
+        bbox_mask = (
             int(float(bboxes[best_idx][0]) * scale_x),
             int(float(bboxes[best_idx][1]) * scale_y),
             int(float(bboxes[best_idx][2]) * scale_x),
             int(float(bboxes[best_idx][3]) * scale_y),
         )
-    else:
-        # Fallback: use bbox as-is (assume same coordinate space)
-        bbox = (
+        bbox_frame = (
             int(float(bboxes[best_idx][0])),
             int(float(bboxes[best_idx][1])),
             int(float(bboxes[best_idx][2])),
             int(float(bboxes[best_idx][3])),
         )
+    else:
+        # Fallback: use bbox as-is for both (assume same coordinate space)
+        bbox_mask = (
+            int(float(bboxes[best_idx][0])),
+            int(float(bboxes[best_idx][1])),
+            int(float(bboxes[best_idx][2])),
+            int(float(bboxes[best_idx][3])),
+        )
+        bbox_frame = bbox_mask
     track_id = int(track_ids[best_idx]) if best_idx < len(track_ids) else best_idx
 
-    return mask, bbox, track_id
+    return mask, bbox_mask, bbox_frame, track_id