Research map / Detection and segmentation / Open-set and few-shot detection
See-Through-Text Grouping for Referring Image Segmentation
Ding-Jie Chen, Songhao Jia, Yi-Chen Lo, Hwann-Tzong Chen, Tyng-Luh Liu
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 pre-trained GloVe embedding and biLSTM (ablation) | STEP model · UNC val | mIoU: 55.11 -> 56.68 | Improved |
| add textual representation learning guided by ConvRNN hidden state (ablation) | STEP model · UNC val | mIoU: 54.40 -> 56.68 | Improved |
| integrate heatmaps with ConvRNN instead of 1x1 convolution (ablation) | STEP model · UNC val | mIoU: 48.71 -> 56.68 | Improved |
| use convLSTM instead of convGRU (ablation) | STEP model · UNC val | mIoU: 55.97 -> 56.68 | Improved |
| use STEP co-embedding instead of concatenation-convolution (ablation) | STEP model · UNC val | mIoU: 48.08 -> 56.68 | Improved |
| initialize first textual representation from biLSTM hidden state instead of uniform heatmap (ablation) | STEP model · UNC val | mIoU: 56.74 -> 56.68 | No clear effect |
| generate STEP heatmaps from low-level to high-level instead of reverse order (ablation) | STEP model · UNC val | mIoU: 57.01 -> 56.68 | No clear effect |
| increase ConvRNN folds from 1 to 4 (ablation) | STEP model · UNC val | mIoU: 56.68 -> 59.13 | Improved |
| increase ConvRNN folds from 1 to 4 (ablation) | STEP model · UNC val | Prec@0.5: 66.03 -> 70.15 | Improved |
| use Ours (1-fold) without DCRF instead of RRN with DCRF | STEP model · UNC val | mIoU: 55.33 -> 56.68 | Improved |
Closest papers by meaning
- Shatter and Gather: Learning Referring Image Segmentation with Text Supervision (ICCV 2023 · similarity 0.91)
- Locate Then Segment: A Strong Pipeline for Referring Image Segmentation (CVPR 2021 · similarity 0.91)
- Boosting Weakly Supervised Referring Image Segmentation via Progressive Comprehension (NeurIPS 2024 · similarity 0.89)
- SetCon: Towards Open-Ended Referring Segmentation via Set-Level Concept Prediction (NeurIPS 2026 · similarity 0.88)
- Mask Grounding for Referring Image Segmentation (CVPR 2024 · similarity 0.88)
- Prompt-Driven Referring Image Segmentation with Instance Contrasting (CVPR 2024 · similarity 0.88)
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.
