Convert boxes and scores to float32 before numpy conversion in process_batch for compatibility

detect-objects.py

@@ -277,7 +277,11 @@ def process_batch(
         image_categories = []
         image_scores = []
 
-        for box, score in zip(boxes.cpu().numpy(), scores.cpu().numpy()):
+        # Convert to float32 before numpy (bfloat16 not supported by numpy)
+        boxes_np = boxes.cpu().float().numpy()
+        scores_np = scores.cpu().float().numpy()
+
+        for box, score in zip(boxes_np, scores_np):
             x1, y1, x2, y2 = box
             width = x2 - x1
             height = y2 - y1