Research map / Vision-language and multimodal / Human and embodied modeling
VisFocus: Prompt-Guided Vision Encoders for OCR-Free Dense Document Understanding
Ofir Abramovich, Niv Nayman, Sharon Fogel, Inbal Lavi, Ron Litman, Shahar Tsiper, Royee Tichauer, Srikar Appalaraju, Shai Mazor, R. Manmatha
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 |
|---|---|---|---|
| replace Swin patch-merging with ViLMA prompt-aware layers (ablation) | VisFocus-B · ChartQA | relaxed accuracy: 52.5 -> 54.7 | Improved |
| replace Swin patch-merging with ViLMA prompt-aware layers (ablation) | VisFocus-B · DocVQA val | ANLS: 70.9 -> 71.3 | Improved |
| add LMPM pre-training stage (ablation) | VisFocus-B · DocVQA val | ANLS: 71.3 -> 71.8 | Improved |
| add LMPM pre-training stage (ablation) | VisFocus-B · ChartQA | relaxed accuracy: 54.7 -> 55.7 | Improved |
| randomly skip concatenating the prompt to the LM input (ablation) | VisFocus-B · ChartQA | relaxed accuracy: 55.7 -> 57.1 | Improved |
| render prompt onto the image (Pix2Struct style) instead of LM-only prompt (ablation) | baseline · DocVQA / ChartQA | ANLS / relaxed accuracy | Hurt |
| use ViLMA in all blocks instead of only the last two (ablation) | VisFocus-B · ChartQA | relaxed accuracy gain over baseline: +2.8 -> +4.6 | Improved |
| use T5-Base encoder rather than independent token embeddings for the prompt (ablation) | VisFocus-B · ChartQA | relaxed accuracy: 55.4 -> 57.1 | Improved |
| use ViLMA prompt guidance on denser documents | VisFocus-B · DocVQA val (documents with at least 400 to 800 words) | ANLS gain over baseline: +0.7 -> +2.3 | Improved |
| apply VisFocus-S over its baseline | VisFocus-S · AI2D | exact match: 42.7 -> 42.6 | No clear effect |
| apply VisFocus-B versus Pix2Struct-B on infographics | VisFocus-B · InfoVQA | ANLS: 38.2 -> 31.9 | Hurt |
Closest papers by meaning
- DocVLM: Make Your VLM an Efficient Reader (CVPR 2025 · similarity 0.88)
- Boosting Document Parsing Efficiency and Performance with Coarse-to-Fine Visual Processing (CVPR 2026 · similarity 0.88)
- DocFormerv2: Local Features for Document Understanding (AAAI 2024 · similarity 0.87)
- LensVLM: Selective Context Expansion for Compressed Visual Representation of Text (NeurIPS 2026 · similarity 0.87)
- Hierarchical Visual Feature Aggregation for OCR-Free Document Understanding (NeurIPS 2024 · similarity 0.87)
- Enhancing Visual Document Understanding with Contrastive Learning in Large Visual-Language Models (CVPR 2024 · similarity 0.87)
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.
