Reproduce “DocXEval: A Systematic Evaluation of Explainability in Multimodal Document Understanding” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Saifullah Saifullah, David Dembinsky, Stefan Agne, Andreas Dengel, Sheraz Ahmed
DocXEval adapts explanation methods and metrics to multimodal document models and tasks with multiple predictions per document. It shares computation across targets where possible, grouping text, layout, position, and image features. Across 13 methods, three tasks, eight datasets, four Transformer models, and 17 metrics, Occlusion is most reliable overall. On FUNSD, its adaptation runs 189.7× faster than iterative batched computation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.