Research map / Detection and segmentation / CNN features
ET-SAM: Efficient Point Prompt Prediction in SAM for Unified Scene Text Detection and Layout Analysis
Xike Zhang, Maoyuan Ye, Juhua Liu, Bo Du
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 Hi-SAM's S-Decoder random 1500 points with the lightweight point decoder (ablation) | Hi-SAM-L · HierText test | average time: 4.4s -> 1.2s with PQ 63.10 -> 63.50 (word) | Improved |
| reduce Hi-SAM random sampled points from 1500 to 280 (ablation) | Hi-SAM-L · HierText test | word PQ: 63.10 -> 61.04 | Hurt |
| add point and mask task prompts together (ablation) | ET-SAM · Total-Text | F-score: 81.99 -> 86.37 | Improved |
| add point and mask task prompts together (ablation) | ET-SAM · HierText layout analysis | F-score: 73.38 -> 74.51 | Improved |
| add a single task prompt alone (ablation) | ET-SAM · CTW1500 | F-score: 86.79 -> 86.30 (mask only), 86.75 (point only), 87.14 (both) | Mixed |
| add Total-Text and CTW1500 to joint training data (ablation) | ET-SAM · Total-Text and CTW1500 | F-score: Total-Text 69.02 -> 85.93 | Improved |
| add Total-Text and CTW1500 to joint training data (ablation) | ET-SAM · HierText test | word PQ: 63.50 -> 61.96 | Hurt |
| raise point threshold to 0.8 (ablation) | ET-SAM · HierText test | word PQ: 62.52 -> 56.98 | Hurt |
| change heatmap loss weight away from 50 (ablation) | ET-SAM · HierText test | word PQ: 62.52 -> 61.61 (25), 62.06 (100) | Hurt |
| fine-tune after joint training | ET-SAM · HierText test | word PQ: 62.52 -> 63.60 | Improved |
| ET-SAM vs Hi-SAM-L in joint training | ET-SAM · Total-Text, CTW1500, ICDAR2015 | F-score: 70.2 -> 87.0 (Total-Text); 77.4 -> 87.9 (CTW1500); 77.4 -> 83.2 (ICDAR2015) | Improved |
| ET-SAM vs specialist detectors | ET-SAM · Total-Text | F-score: 87.0 vs 90.0 (DeepSolo) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- ESTextSpotter — ESTextSpotter: Towards Better Scene Text Spotting with Explicit Synergy in Transformer (ICCV 2023) · better on 0 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Bridge the Points: Graph-based Few-shot Segment Anything Semantically (NeurIPS 2024 · similarity 0.88)
- Segment and Caption Anything (CVPR 2024 · similarity 0.87)
- SAM-CP: Marrying SAM with Composable Prompts for Versatile Segmentation (ICLR 2025 · similarity 0.87)
- Stable Segment Anything Model (ICLR 2025 · similarity 0.87)
- AoP-SAM: Automation of Prompts for Efficient Segmentation (AAAI 2025 · similarity 0.87)
- LayoutFormer: Hierarchical Text Detection Towards Scene Text Understanding (CVPR 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.
