Research map / Vision-language and multimodal / Human and embodied modeling
Decoupling What to Count and Where to See for Referring Expression Counting
Yuda Zou, Zijian Zhang, Yongchao Xu
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 Subclass Separable Matching (SSM) (ablation) | baseline · REC-8K val (Swin-T) | MAE: 7.21 -> 5.93 | Improved |
| add W2 Decoder (ablation) | baseline · REC-8K val (Swin-T) | MAE: 7.21 -> 3.85 | Improved |
| add SSM on top of W2 Decoder (ablation) | W2-Net · REC-8K val (Swin-T) | MAE: 3.85 -> 3.55 | Improved |
| add W2 Decoder (ablation) | baseline · REC-8K val (Swin-T) | F1: 0.66 -> 0.76 | Improved |
| raise repulsive weight from 0 to 0.2 (ablation) | SSM · REC-8K val (Swin-T) | MAE: 3.85 -> 3.55 | Improved |
| raise repulsive weight from 0.2 to 1.0 (ablation) | SSM · REC-8K val (Swin-T) | MAE: 3.55 -> 3.92 | Hurt |
| use exponential ratio repulsive cost instead of hinge on ratio (ablation) | SSM · REC-8K val (Swin-T, no W2 Decoder) | MAE: 6.41 -> 5.93 | Improved |
| use exponential ratio cost instead of inverse negative distance (ablation) | SSM · REC-8K val (Swin-T, no W2 Decoder) | MAE: 7.05 -> 5.93 | Improved |
| use exponential ratio cost instead of normalized positive distance (ablation) | SSM · REC-8K val (Swin-T, no W2 Decoder) | MAE: 6.88 -> 5.93 | Improved |
| use W2-Net instead of CAD-GD (density query), Swin-T | REC counting · REC-8K val | MAE: 4.58 -> 3.55 | Improved |
| use W2-Net instead of CAD-GD (density query), Swin-T | REC counting · REC-8K test | MAE: 4.59 -> 3.59 | Improved |
| use W2-Net instead of CAD-GD (text) on zero-shot counting | FSC-147 · FSC-147 val | MAE: 9.30 -> 8.73 | Improved |
Closest papers by meaning
- Referring Expression Counting (CVPR 2024 · similarity 0.92)
- Exploring Contextual Attribute Density in Referring Expression Counting (CVPR 2025 · similarity 0.90)
- Revisiting Counterfactual Problems in Referring Expression Comprehension (CVPR 2024 · similarity 0.86)
- Rethinking Two-Stage Referring Expression Comprehension: A Novel Grounding and Segmentation Method Modulated by Point (AAAI 2024 · similarity 0.85)
- OmniCount: Multi-label Object Counting with Semantic-Geometric Priors (AAAI 2025 · similarity 0.85)
- A Fixed-Point Approach to Unified Prompt-Based Counting (AAAI 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.
