Efficient End-to-End Visual Document Understanding with Rationale Distillation
This addresses efficiency challenges for applications requiring visual document analysis, but it is incremental as it builds on existing methods like OCR and LLMs.
The paper tackles the problem of high computational and engineering complexity in visual document understanding by proposing Rationale Distillation, which trains a small student model using intermediate rationales from larger tools, resulting in a 4-5% absolute accuracy improvement on benchmarks with only a 1% higher computational cost.
Understanding visually situated language requires interpreting complex layouts of textual and visual elements. Pre-processing tools, such as optical character recognition (OCR), can map document image inputs to textual tokens, then large language models (LLMs) can reason over text. However, such methods have high computational and engineering complexity. Can small pretrained image-to-text models accurately understand visual documents through similar recognition and reasoning steps instead? We propose Rationale Distillation (RD), which incorporates the outputs of OCR tools, LLMs, and larger multimodal models as intermediate "rationales", and trains a small student model to predict both rationales and answers. On three visual document understanding benchmarks representing infographics, scanned documents, and figures, our Pix2Struct (282M parameters) student model finetuned with RD outperforms the base model by 4-5% absolute accuracy with only 1% higher computational cost.