Research map / Vision-language and multimodal / CLIP and zero-shot
Decomposed Soft Prompt Guided Fusion Enhancing for Compositional Zero-Shot Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 CSP with DFSP using t2i fusion | CLIP-based CZSL · MIT-States closed-world | AUC: 19.4 -> 20.6 | Helped |
| Replace CSP with DFSP using t2i fusion | CLIP-based CZSL · UT-Zappos closed-world | AUC: 33.0 -> 36.0 | Helped |
| Replace CSP with DFSP using t2i fusion | CLIP-based CZSL · C-GQA closed-world | AUC: 6.2 -> 10.5 | Helped |
| Add self-attention to SPM only (ablation) | DFSP soft prompt model · UT-Zappos closed-world | AUC: 33.1 -> 32.6 | Hurt |
| Add self-attention to SPM only (ablation) | DFSP soft prompt model · MIT-States closed-world | AUC: 19.1 -> 20.6 | Helped |
| Add fusion without decomposition (ablation) | DFSP soft prompt model · MIT-States and UT-Zappos closed-world | AUC: 19.1 -> 18.3; 33.1 -> 21.4 | Hurt |
| Add language-feature decomposition without fusion (ablation) | DFSP soft prompt model · MIT-States and UT-Zappos closed-world | AUC: 19.1 -> 20.7; 33.1 -> 33.5 | Helped |
| Add decomposed fusion module (ablation) | DFSP soft prompt model · MIT-States and UT-Zappos closed-world | AUC: 19.1 -> 20.6; 33.1 -> 36.0 | Mixed |
| Add self-attention to SPM only (ablation) | DFSP soft prompt model · MIT-States and UT-Zappos open-world | AUC: 6.1 -> 6.6; 23.5 -> 26.9 | Helped |
| Add language-feature decomposition without fusion (ablation) | DFSP soft prompt model · MIT-States and UT-Zappos open-world | AUC: 6.1 -> 6.6; 23.5 -> 26.4 | Mixed |
| Add decomposed fusion module (ablation) | DFSP soft prompt model · MIT-States and UT-Zappos open-world | AUC: 6.1 -> 6.8; 23.5 -> 30.3 | Helped |
| Fuse decomposed text features into the image branch (t2i) rather than the text branch (i2t) (ablation) | DFSP · Closed-world benchmarks | AUC: MIT-States: 20.7 -> 20.6; UT-Zappos: 32.1 -> 36.0; C-GQA: 8.7 -> 10.5 | Mixed |
Closest papers by meaning
- Visual and Semantic Prompt Collaboration for Generalized Zero-Shot Learning (CVPR 2025 · similarity 0.82)
- Attend and Enrich: Enhanced Visual Prompt for Zero-Shot Learning (AAAI 2025 · similarity 0.81)
- ViLT-CLIP: Video and Language Tuning CLIP with Multimodal Prompt Learning and Scenario-Guided Optimization (AAAI 2024 · similarity 0.81)
- Revealing the Proximate Long-Tail Distribution in Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.79)
- Beyond Seen Primitive Concepts and Attribute-Object Compositional Learning (CVPR 2024 · similarity 0.79)
- On the Test-Time Zero-Shot Generalization of Vision-Language Models: Do We Really Need Prompt Learning? (CVPR 2024 · similarity 0.79)
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.