Research map / Vision-language and multimodal / CLIP and zero-shot
Learning complete and explainable visual representations from itemized text supervision
Yiwei Lyu, Chenhui Zhao, Soumyanil Banerjee, Shixuan Liu, Akshay Rao, Akhil Kondepudi, Honglak Lee, Todd C. Hollon
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 Inter-Item Separation (IIS) loss (ablation) | TCS · Pub-Brain-5 | 12-metric mean BAcc: 79.9 -> 81.2 | Improved |
| add IIS loss (ablation) | TCS · Brain MRI | mean attention map similarity (mAMS, lower better): 0.50 -> 0.33 (TCS+MPS vs +IIS) | Improved |
| add IIS loss (ablation) | TCS · BraTS2021 zero-shot segmentation | mIoU: 9.1 -> 6.5 | Hurt |
| add Multi-Positive SigLIP (MPS) on top of TCS+IIS (ablation) | ItemizedCLIP · BraTS2021 zero-shot segmentation | mIoU: 6.5 -> 16.1 | Improved |
| add MPS (ablation) | TCS+IIS · Brain MRI completeness | MLL x100: 40.29 -> 39.43 | Hurt |
| add Upweight Worst Positive (UWP) (ablation) | TCS+IIS+MPS · Brain MRI completeness | MLL x100: 39.43 -> 42.96 | Improved |
| add UWP (ablation) | TCS+IIS+MPS · Pub-Brain-5 | 12-metric mean BAcc: 81.9 -> 81.6 | No clear effect |
| add Key Token Alignment (KTA) (ablation) | TCS+IIS+MPS+UWP · RSICD | top-1 accuracy: 42.9 -> 44.6 | Improved |
| add KTA (ablation) | TCS+IIS+MPS+UWP · Brain MRI completeness | MLL x100: 42.96 -> 42.46 | Hurt |
| add masked cross-attention in TCS (full ItemizedCLIP) (ablation) | TCS+IIS+MPS+UWP+KTA · RSICD | top-1 accuracy: 44.6 -> 46.2 | Improved |
| use ItemizedCLIP objective instead of FLAIR TCS+MPS (ablation) | FLAIR · Pub-Brain-5 | 12-metric mean BAcc: 78.7 -> 83.6 | Improved |
| use ItemizedCLIP instead of HLIP | HLIP · UM220K brain MRI | mean AUC: 83.7 -> 90.5 | Improved |
Closest papers by meaning
- SuperCLIP: CLIP with Simple Classification Supervision (NeurIPS 2025 · similarity 0.86)
- SmartCLIP: Modular Vision-language Alignment with Identification Guarantees (CVPR 2025 · similarity 0.85)
- FineCLIP: Self-distilled Region-based CLIP for Better Fine-grained Understanding (NeurIPS 2024 · similarity 0.85)
- UVU: Improving Multimodal Understanding via Vision-Language Unified Autoregressive Paradigm (CVPR 2026 · similarity 0.84)
- Learning Interleaved Image-Text Comprehension in Vision-Language Large Models (ICLR 2025 · similarity 0.84)
- Enhancing Descriptive Captions with Visual Attributes for Multimodal Perception (CVPR 2026 · similarity 0.84)
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.
