Research map / Vision-language and multimodal / CLIP and zero-shot
Prompting Language-Informed Distribution for Compositional Zero-Shot Learning
Wentao Bao, Lichang Chen, Heng Huang, Yu Kong
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 language-informed distribution (LID) over LLM descriptions (ablation) | PLID baseline · MIT-States (closed-world) | AUC: 18.56 -> 20.43 | Improved |
| add text and visual feature enhancement (ablation) | PLID · MIT-States (closed-world) | AUC: 20.43 -> 21.09 | Improved |
| add primitive decomposition for fused decision (ablation) | PLID · MIT-States (closed-world) | AUC: 21.67 -> 22.12 | Improved |
| use larger LLMs (GPT-3.5, Mistral-7B) instead of OPT-1.3B for descriptions (ablation) | PLID · MIT-States (closed-world) | AUC: 22.12 -> 20.61 (GPT-3.5), 20.78 (Mistral-7B) | Hurt |
| use 3 cross-attention layers instead of 1 in TFE and VFE (ablation) | PLID · MIT-States (closed-world) | AUC: 22.12 -> 20.65 | Hurt |
| use stochastic instead of deterministic logit fusion (ablation) | PLID · MIT-States (closed-world) | AUC: 21.90 -> 22.12 | Improved |
| decompose both visual and text features rather than text only (ablation) | PLID · MIT-States (closed-world) | AUC: 21.31 -> 21.69 | Improved |
| apply LID on states and objects in addition to compositions (ablation) | PLID · MIT-States (closed-world) | AUC: 21.90 -> 22.12 | Improved |
| apply LID on states and objects only (no composition level) (ablation) | PLID · MIT-States (closed-world) | AUC: 21.67 -> 21.62 | No clear effect |
| share covariance across attributes of the same object (ablation) | PLID · MIT-States (closed-world) | AUC: 22.12 -> 21.69; memory 22.2 -> 17.6 GB | Hurt |
| PLID with LLM-informed distributions vs ProDA learned prompt distributions | CLIP-based CZSL · MIT-States (closed-world) | AUC: 16.1 -> 20.43 | Improved |
Closest papers by meaning
- Beyond Seen Primitive Concepts and Attribute-Object Compositional Learning (CVPR 2024 · similarity 0.91)
- A Dynamic Learning Method towards Realistic Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.91)
- Revealing the Proximate Long-Tail Distribution in Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.91)
- LOGICZSL: Exploring Logic-induced Representation for Compositional Zero-shot Learning (CVPR 2025 · similarity 0.90)
- Context-based and Diversity-driven Specificity in Compositional Zero-Shot Learning (CVPR 2024 · similarity 0.90)
- Retrieval-Augmented Primitive Representations for Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.90)
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.
