Research map / Label-efficient and robust learning / Recognition and long-tail
Evaluating Out-of-Distribution Performance on Document Image Classifiers
Stefan Larson, Gordon Lim, Yutong Ai, David Kuang, Kevin Leach
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| evaluate on RVL-CDIP-N instead of RVL-CDIP test | DiT-base · RVL-CDIP vs RVL-CDIP-N | micro accuracy: 0.933 -> 0.786 | Hurt |
| evaluate on RVL-CDIP-N instead of RVL-CDIP test | VGG-16 · RVL-CDIP vs RVL-CDIP-N | micro accuracy: 0.905 -> 0.668 | Hurt |
| evaluate on RVL-CDIP-N instead of RVL-CDIP test | LayoutLMv2 · RVL-CDIP vs RVL-CDIP-N | micro accuracy: 0.887 -> 0.556 | Hurt |
| evaluate on RVL-CDIP-N instead of RVL-CDIP test | AlexNet · RVL-CDIP vs RVL-CDIP-N | micro accuracy: 0.885 -> 0.567 | Hurt |
| add Augraphy scanner noise to RVL-CDIP-N test images (ablation) | DiT-base · RVL-CDIP-N | micro accuracy: 0.786 -> 0.706 | Hurt |
| add Augraphy scanner noise to RVL-CDIP-N test images (ablation) | VGG-16 · RVL-CDIP-N | micro accuracy: 0.668 -> 0.650 | No clear effect |
| add Augraphy scanner noise to RVL-CDIP-N test images (ablation) | ResNet-50 · RVL-CDIP-N | micro accuracy: 0.598 -> 0.606 | No clear effect |
| switch in-domain set from RVL-CDIP test to RVL-CDIP-N for OOD detection | DiT-base · RVL-CDIP-O detection | AUC (MSP): 0.894 -> 0.728 | Hurt |
| switch in-domain set from RVL-CDIP test to RVL-CDIP-N for OOD detection | DiT-base · RVL-CDIP-O detection | AUC (Energy): 0.888 -> 0.753 | Hurt |
| switch in-domain set from RVL-CDIP test to RVL-CDIP-N for OOD detection | VGG-16 · RVL-CDIP-O detection | AUC (Energy): 0.923 -> 0.648 | Hurt |
| use Energy instead of MSP confidence scoring | DiT-base · RVL-CDIP-N vs RVL-CDIP-O | AUC: 0.728 -> 0.753 | Improved |
| use Energy instead of MSP confidence scoring | ResNet-50 · RVL-CDIP-N vs RVL-CDIP-O | AUC: 0.581 -> 0.554 | Hurt |
Closest papers by meaning
- OmniDocBench: Benchmarking Diverse PDF Document Parsing with Comprehensive Annotations (CVPR 2025 · similarity 0.84)
- MDPBench: A Benchmark for Multilingual Document Parsing in Real-World Scenarios (NeurIPS 2026 · similarity 0.83)
- Unsupervised Learning for Class Distribution Mismatch (ICML 2025 · similarity 0.83)
- Towards Natural Language-Based Document Image Retrieval: New Dataset and Benchmark (CVPR 2025 · similarity 0.82)
- Generative Classifiers Avoid Shortcut Solutions (ICLR 2025 · similarity 0.82)
- LCA-on-the-Line: Benchmarking Out of Distribution Generalization with Class Taxonomies (ICML 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
