Research map / Detection and segmentation / Open-set and few-shot detection
Hi-SAM: Marrying Segment Anything Model for Hierarchical Text Segmentation
Maoyuan Ye, Jing Zhang, Juhua Liu, Chenyu Liu, Baocai Yin, Cong Liu, Bo Du, Dacheng Tao
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 adapters in the SAM backbone (ablation) | SAM-TS-B · Total-Text | fgIOU: 68.80 -> 79.73 | Improved |
| add adapters in the SAM backbone (ablation) | SAM-TS-L · Total-Text | fgIOU: 69.17 -> 84.02 | Improved |
| add Transformer decoder layer in self-prompting (ablation) | SAM-TS-L · Total-Text | fgIOU: 84.02 -> 84.59 | Improved |
| use spatial attention tokens instead of vanilla embedding (ablation) | SAM-TS-L · Total-Text | fgIOU: 83.92 -> 84.59 | Improved |
| add high-resolution mask features in S-Decoder (ablation) | SAM-TS-L · HierText | fgIOU: 64.13 -> 78.37 | Improved |
| add high-resolution mask features in S-Decoder (ablation) | SAM-TS-L · Total-Text | fgIOU: 82.28 -> 84.59 | Improved |
| use adapter-tuning instead of model-tuning (ablation) | SAM-TS-L · Total-Text | fgIOU: 80.52 -> 84.59 | Improved |
| use SAM weights instead of MAE pretrained ViT for initialization (ablation) | SAM-TS-L · Total-Text | fgIOU: 75.39 -> 80.52 | Improved |
| increase prompt tokens from 12 to 16 (ablation) | SAM-TS-L · Total-Text | fgIOU: 84.59 -> 83.94 | Hurt |
| increase word mask feature size from 256 to 384 (ablation) | Hi-SAM-L · HierText val | word PQ: 60.15 -> 62.60 | Improved |
| increase points per text-line from 2 to 5 during training (ablation) | Hi-SAM-L · HierText val | word PQ: 62.60 -> 63.14 | Improved |
| increase sampled text-lines from 10 to 25 (ablation) | Hi-SAM-L · HierText val | layout analysis PQ: 58.83 -> 60.03 | Improved |
Closest papers by meaning
- SAM-CP: Marrying SAM with Composable Prompts for Versatile Segmentation (ICLR 2025 · similarity 0.87)
- Exploring Simple Open-Vocabulary Semantic Segmentation (CVPR 2025 · similarity 0.87)
- LayoutFormer: Hierarchical Text Detection Towards Scene Text Understanding (CVPR 2024 · similarity 0.86)
- SRFormer: Text Detection Transformer with Incorporated Segmentation and Regression (AAAI 2024 · similarity 0.86)
- From SAM to CAMs: Exploring Segment Anything Model for Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.86)
- Bridge the Points: Graph-based Few-shot Segment Anything Semantically (NeurIPS 2024 · 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.
