VisFocus: Prompt-Guided Vision Encoders for OCR-Free Dense Document Understanding Paper • 2407.12594 • Published Jul 17 • 19