Research map / Architectures and efficiency / Attention and transformers
DREAM: Document Recognition with Explicit Adaptive Memory
Tianqi Zhao, Di Wu, Liangrui Peng, Yifan Huang, Kemeng Zhao, Shuo Li, Zhiyu Li, Yizhu Wang, Borui Jiang, Yuyang Li
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 |
|---|---|---|---|
| add multiscale prototype memory with sparsity loss | GOT-based document recognizer · Fox (English) | Edit Distance: 0.107 -> 0.082 | Improved |
| add multiscale prototype memory with sparsity loss | GOT-based document recognizer · Fox (Chinese) | Edit Distance: 0.112 -> 0.101 | Improved |
| add DREAM memory | GOT-based document recognizer · DreamDoc test | Edit Distance: 0.254 -> 0.248 | Improved |
| add DREAM memory | GOT-based document recognizer · DreamDoc test | F1-score: 0.868 -> 0.869 | No clear effect |
| add DREAM memory to handwriting model | handwriting text line recognizer · SCUT-HCCDoc | AR (%): 92.78 -> 93.36 | Improved |
| add DREAM memory to handwriting model | handwriting text line recognizer · SCUT-EPT | AR (%): 78.48 -> 78.82 | Improved |
| use multiscale memory instead of single-scale memory (ablation) | DREAM · Fox (Chinese) | Edit Distance: 0.126 -> 0.101 | Improved |
| add entropy sparsity loss to multiscale memory (ablation) | DREAM · Fox (Chinese) | Edit Distance: 0.104 -> 0.101 | Improved |
| add entropy sparsity loss to multiscale memory (ablation) | DREAM · Fox (English) | Edit Distance: 0.096 -> 0.082 | Improved |
| add global negative entropy term to local entropy loss (ablation) | DREAM · Fox (Chinese) | Edit Distance: 0.136 -> 0.101 | Improved |
| add global negative entropy term to local entropy loss (ablation) | DREAM · Fox (English) | Edit Distance: 0.087 -> 0.082 | Improved |
| use single-scale memory with sparsity loss instead of no memory (ablation) | GOT-based document recognizer · Fox (Chinese) | Edit Distance: 0.112 -> 0.126 | Hurt |
Closest papers by meaning
- Recurrence-Enhanced Vision-and-Language Transformers for Robust Multimodal Document Retrieval (CVPR 2025 · similarity 0.86)
- Boosting Document Parsing Efficiency and Performance with Coarse-to-Fine Visual Processing (CVPR 2026 · similarity 0.84)
- Your Embedding Model Is SMARTer Than You Think (NeurIPS 2026 · similarity 0.84)
- DocLayLLM: An Efficient Multi-modal Extension of Large Language Models for Text-rich Document Understanding (CVPR 2025 · similarity 0.84)
- Document Haystacks: Vision-Language Reasoning Over Piles of 1000+ Documents (CVPR 2025 · similarity 0.83)
- Towards Real-World Document Parsing via Realistic Scene Synthesis and Document-Aware Training (CVPR 2026 · similarity 0.83)
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.
