Research map / Detection and segmentation / Open-set and few-shot detection
DocLayout-VL: A Foundational Model for Hierarchical, Open-set, and Promptable Document Layout Segmentation
Venkata Venna, Srihari Bandarupalli, Anirudh Srinivasan, R Raghuveer, Sai Madhusudan Gunda, SANTOSH RAVI KIRAN SARVADEVABHATLA
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 flat bbox-trained fine-tuned VLM with DocLayout-VL segmentation and hierarchy design | Qwen3-VL-8B · HOPLay-Docs-Bench | Overall Score: 45.8 -> 79.4 | Improved |
| add Hierarchical Grounding Contextualizer (ablation) | DocLayout-VL · HierLay-Docs-Bench | Open mIoU: 58.3 -> 77.5 | Improved |
| add SAM segmentation module and HGC to base VLM (ablation) | Qwen3-VL-8B · HierLay-Docs-Bench | Open mIoU: 32.5 -> 77.5 | Improved |
| add structural loss L_struct (ablation) | DocLayout-VL · HierLay-Docs-Bench | Open mIoU: 72.3 -> 77.5 | Improved |
| add background exclusion loss L_bg (ablation) | DocLayout-VL · HierLay-Docs-Bench | Open mIoU: 69.2 -> 72.3 | Improved |
| use Qwen3-VL-8B backbone instead of InternVL3.5-8B (ablation) | DocLayout-VL · HierLay-Docs-Bench | Open mIoU: 71.3 -> 77.5 | Improved |
| use three-stage curriculum instead of single joint fine-tuning (ablation) | DocLayout-VL · HierLay-Docs-Bench (100% data) | Open mIoU: 68.7 -> 77.5 | Improved |
| add HGC versus doubling data volume (ablation) | DocLayout-VL · HierLay-Docs-Bench | Open mIoU: VLM+SAM 47.6 -> 53.1 with 25% -> 50% data; +HGC at 25% data 59.8 | Improved |
| scale VLM size within a family (Qwen3-VL-8B to 235B) | zero-shot VLMs · HOPLay-Docs-Bench | Overall Score: 31.6 -> 31.5 | No clear effect |
| zero-shot DocLayout-VL versus per-dataset trained detector on PubLayNet | DocLayout-VL-8B-it · PubLayNet | mAP@50:95: 96.9 (DocLayout-YOLO) vs 96.5 | No clear effect |
| zero-shot DocLayout-VL versus DocLayout-YOLO on label-rich benchmark | DocLayout-VL-8B-it · IndicDLP | mAP@50:95: 54.5 -> 77.5 | Improved |
Closest papers by meaning
- LayoutLLM: Layout Instruction Tuning with Large Language Models for Document Understanding (CVPR 2024 · similarity 0.89)
- A Simple yet Effective Layout Token in Large Language Models for Document Understanding (CVPR 2025 · similarity 0.89)
- M2Doc: A Multi-Modal Fusion Approach for Document Layout Analysis (AAAI 2024 · similarity 0.88)
- Graph-based Document Structure Analysis (ICLR 2025 · similarity 0.87)
- Seeing is Improving: Visual Feedback for Iterative Text Layout Refinement (CVPR 2026 · similarity 0.86)
- DocLayLLM: An Efficient Multi-modal Extension of Large Language Models for Text-rich Document Understanding (CVPR 2025 · similarity 0.86)
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.
