Research map / Detection and segmentation / CNN features
I3CL: Intra- and Inter-Instance Collaborative Learning for Arbitrary-Shaped Scene Text Detection
Bo Du, Jian Ye, Jing Zhang, Juhua Liu, Dacheng Tao
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 Intra-CL module (ablation) | Mask R-CNN baseline · ArT | F-measure: 74.0 -> 75.0 | Improved |
| add Intra-CL module (ablation) | Mask R-CNN baseline · Total-Text | F-measure: 83.3 -> 84.7 | Improved |
| add Intra-CL module (ablation) | Mask R-CNN baseline · CTW-1500 | F-measure: 82.8 -> 84.1 | Improved |
| add Inter-CL (GCG) on top of Intra-CL (ablation) | Intra-CL · ArT | F-measure: 75.0 -> 76.6 | Improved |
| use transformer (GCT) instead of average pooling (GCG) for Inter-CL context (ablation) | Intra-CL + Inter-CL · ArT | F-measure: 76.6 -> 76.4 | No clear effect |
| combine GCG and GCT in Inter-CL (ablation) | Intra-CL + Inter-CL · ArT | F-measure: 76.6 -> 76.5 | No clear effect |
| add semi-supervised learning with pseudo labels (ablation) | I3CL · ArT | F-measure: 76.6 -> 77.5 | Improved |
| use 3-path instead of 1-path Intra-CL (ablation) | Intra-CL · ArT | F-measure: 74.4 -> 75.0 | Improved |
| reverse kernel order from 753 to 357 (ablation) | Intra-CL · ArT | F-measure: 75.0 -> 74.3 | Hurt |
| concatenation vs element-wise sum for path fusion (ablation) | Intra-CL · ArT | F-measure: 74.9 -> 75.0 | No clear effect |
| use threshold filter pseudo labels from a single model (ablation) | I3CL baseline · ArT | F-measure: 76.6 -> 75.5 | Hurt |
| use ensemble with Overlap-Mask and Soft-Box for pseudo labels (ablation) | I3CL · ArT | F-measure: 76.6 -> 77.5 | Improved |
Later papers that compared with this method
1 later accepted paper reports I3CL as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- DPText-DETR: Towards Better Scene Text Detection with Dynamic Points in Transformer (AAAI 2023 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ContourNet — ContourNet: Taking a Further Step Toward Accurate Arbitrary-Shaped Scene Text Detection (CVPR 2020) · better on 1 of 1 shared result
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
- EAST: An Efficient and Accurate Scene Text Detector (CVPR 2017 · similarity 0.87)
- Learning Shape-Aware Embedding for Scene Text Detection (CVPR 2019 · similarity 0.87)
- Towards Unconstrained End-to-End Text Spotting (ICCV 2019 · similarity 0.87)
- Mask TextSpotter v3: Segmentation Proposal Network for Robust Scene Text Spotting (ECCV 2020 · similarity 0.87)
- InstructOCR: Instruction Boosting Scene Text Spotting (AAAI 2025 · similarity 0.85)
- Towards Unified Multi-granularity Text Detection with Interactive Attention (ICML 2024 · similarity 0.85)
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.
