Visual Attention based OCR. The model first runs a sliding CNN on the image (images are resized to height 32 while preserving aspect ratio).
確定！ 回上一頁
