Research map / Vision-language and multimodal / CLIP and zero-shot
What is the Right Embedding Space for Contrastive Learning in Referring Expression Counting?
Kostas Triaridis, Panagiotis Kaliosis, E-Ro Nguyen, Jingyi Xu, Dimitris Samaras, Hieu Le
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 image-text contrastive loss with image-space contrastive loss (ablation) | GroundingREC · REC-8K val | MAE: 6.80 -> 4.74 | Improved |
| replace image-text contrastive loss with image-space contrastive loss (ablation) | CAD-GD · REC-8K val | MAE: 6.33 -> 5.98 | Improved |
| add image-space CL on top of image-text CL (ablation) | GroundingREC · REC-8K val | MAE: 6.80 -> 5.39 | Improved |
| add image-text CL alongside image-space CL (ablation) | GroundingREC · REC-8K val | MAE: 4.74 -> 5.39 | Hurt |
| use Bipartite Matching instead of top-N positives (ablation) | GroundingDINO + C-REX · REC-8K val | MAE: 4.86 -> 4.26 | Improved |
| use Bipartite Matching instead of top-5 positives (ablation) | GroundingDINO + C-REX · REC-8K val | MAE: 7.58 -> 4.26 | Improved |
| use top N instead of top N+sqrt(N) positives (ablation) | GroundingDINO + C-REX · REC-8K val | MAE: 6.02 -> 4.86 | Improved |
| add standard supervised contrastive loss (ablation) | GroundingDINO · REC-8K test | MAE: 5.90 -> 5.91 | No clear effect |
| use modified loss with only positive samples as anchors (ablation) | GroundingDINO · REC-8K val | MAE: 5.69 -> 4.86 | Improved |
| add C-REX | CAD-GD · REC-8K test (paraphrased REs) | MAE: 6.85 -> 8.02 | Hurt |
| add C-REX | GrREC · REC-8K test (paraphrased REs) | MAE: 9.88 -> 6.22 | Improved |
| add C-REX | GroundingREC · FSC-147 test | MAE: 10.12 -> 12.46 | Hurt |
Closest papers by meaning
- Exploring Contextual Attribute Density in Referring Expression Counting (CVPR 2025 · similarity 0.88)
- Revisiting Counterfactual Problems in Referring Expression Comprehension (CVPR 2024 · similarity 0.88)
- Referring Expression Counting (CVPR 2024 · similarity 0.87)
- RefDetector: A Simple Yet Effective Matching-based Method for Referring Expression Comprehension (AAAI 2025 · similarity 0.87)
- T2ICount: Enhancing Cross-modal Understanding for Zero-Shot Counting (CVPR 2025 · similarity 0.87)
- VLCounter: Text-Aware Visual Representation for Zero-Shot Object Counting (AAAI 2024 · similarity 0.87)
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.
