A document processing team uses DLPy object detection to identify and localize table regions in scanned financial report images, requiring bounding box coordinates. Which architecture is most appropriate?

B is correct because table detection with bounding box output is an object detection problem requiring architectures that predict both class labels and spatial coordinates (x, y, width, height) for multiple regions within an image. DLPy supports YOLO-style architectures designed for this. A is wrong; patch-based classifiers cannot efficiently produce bounding box coordinates for the full page. C is wrong; text classifiers operate on text sequences, not images. D is wrong; sequence-to-sequence models are for text translation/generation, not spatial bounding box regression.