Enhance batch processing to set tensor dtype for model compatibility

detect-objects.py

@@ -236,7 +236,7 @@ def process_batch(
             images=pil_images,
             text=[class_name] * len(pil_images),  # Same prompt for all images
             return_tensors="pt",
-        ).to(model.device)
+        ).to(model.device, dtype=model.dtype)
 
         with torch.no_grad():
             outputs = model(**inputs)