Research map / Vision-language and multimodal / CLIP and zero-shot
SynCLIP: Synonym-Coherent Language-Image Pretraining for Robust Open-Vocabulary Dense Perception
Mingjie Xie, Guangjun He, Dongli Xu, Youtian Lin, Hongjue Li, Pengming Feng, Jian Guan, Yue Deng
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 SSA and SAR together (ablation) | F-ViT (DeCLIP baseline) · OV-COCO | AP50 novel: 41.0 -> 43.6 | Improved |
| add SSA alone (ablation) | F-ViT baseline · OV-COCO | AP50 novel: 41.0 -> 41.1 | No clear effect |
| add SAR alone (ablation) | F-ViT baseline · OV-COCO | AP50 novel: 41.0 -> 42.2 | Improved |
| add SSA on top of SAR (ablation) | SynCLIP · OV-COCO | AP50 novel: 42.2 -> 43.6 | Improved |
| add synonyms as SSA reference text (ablation) | SynCLIP · OV-COCO | AP50 novel: 41.1 -> 42.1 | Improved |
| add definitions as SSA reference text (ablation) | SynCLIP · OV-COCO | AP50 novel: 41.1 -> 42.3 | Improved |
| use both synonyms and definitions (ablation) | SynCLIP · OV-COCO | AP50 novel: 41.1 -> 43.6 | Improved |
| balance alpha and beta at 0.5 instead of alpha 0.9, beta 0.1 (ablation) | SAR aggregation · OV-COCO | AP50 novel: 41.4 -> 43.6 | Improved |
| balance alpha and beta at 0.5 instead of alpha 0.1, beta 0.9 (ablation) | SAR aggregation · OV-COCO | AP50 novel: 41.7 -> 43.6 | Improved |
| increase top-k tokens from 1 to 7 (ablation) | SAR token selection · OV-COCO | AP50 novel: 42.5 -> 43.6 | Improved |
| increase top-k tokens from 7 to 20 (ablation) | SAR token selection · OV-COCO | AP50 novel: 43.6 -> 41.5 | Hurt |
| replace SynCLIP pretraining with CLIPSelf under synonym substitution | F-ViT · OV-COCO (synonym substitution) | AP50 novel: 37.6 -> 28.8 (-8.8) | Hurt |
Closest papers by meaning
- Refining CLIP's Spatial Awareness: A Visual-Centric Perspective (ICLR 2025 · similarity 0.90)
- DeCLIP: Decoupled Learning for Open-Vocabulary Dense Perception (CVPR 2025 · similarity 0.89)
- ResCLIP: Residual Attention for Training-free Dense Vision-language Inference (CVPR 2025 · similarity 0.88)
- GroundingME: Exposing the Visual Grounding Gap in MLLMs through Multi-Dimensional Evaluation (CVPR 2026 · similarity 0.88)
- MMLongCite: A Benchmark for Evaluating Faithfulness of Long-Context Vision-Language Models (NeurIPS 2026 · similarity 0.88)
- Your Large Vision-Language Model Only Needs A Few Attention Heads For Visual Grounding (CVPR 2025 · 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.
