Research map / Vision-language and multimodal / CLIP and zero-shot
DICE: Disentangled Instance-Class knowlEdge prompt tuning via SAE for Vision-Language Models
Daeun Lee, Seungwoo Jang, Kwangsu Kim
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 instance context module to class context module (ablation) | DICE · Base-to-novel ablation (11 datasets avg) | harmonic mean: 79.80 -> 80.43 | Improved |
| add class context module to instance context module (ablation) | DICE · Base-to-novel ablation (11 datasets avg) | harmonic mean: 78.22 -> 80.43 | Improved |
| use SAE instead of same-size MLP (ablation) | DICE · Base-to-novel ablation (11 datasets avg) | harmonic mean: 77.38 -> 80.43 | Improved |
| use LLM embeddings instead of CLIP text embeddings (ablation) | DICE · Base-to-novel ablation (11 datasets avg) | harmonic mean: 79.56 -> 80.43 | Improved |
| use text-embedding-3-large instead of ada-002 (ablation) | DICE · Base-to-novel ablation (11 datasets avg) | harmonic mean: 78.92 -> 80.43 | Improved |
| use text-embedding-3-large instead of 3-small (ablation) | DICE · Base-to-novel ablation (11 datasets avg) | harmonic mean: 79.95 -> 80.43 | Improved |
| select k=4 concepts instead of 2 (ablation) | DICE · Base-to-novel ablation (11 datasets avg) | harmonic mean: 79.93 -> 80.43 | Improved |
| select k=4 concepts instead of 8 (ablation) | DICE · Base-to-novel ablation (11 datasets avg) | harmonic mean: 80.12 -> 80.43 | Improved |
| vary SAE top-k and expansion ratio (ablation) | DICE · Base-to-novel ablation (11 datasets avg) | harmonic mean: 79.73 to 80.43 | No clear effect |
| vary lambda_kg among 6, 8, 10 (ablation) | DICE · 11-dataset few-shot | average accuracy: 79.37, 79.46, 79.23 | No clear effect |
| plug DICE into CoOp | CoOp · Base-to-novel (11 datasets avg, 16-shot) | harmonic mean: 71.66 -> 77.45 | Improved |
| plug DICE into CoCoOp | CoCoOp · Base-to-novel (11 datasets avg, 16-shot) | harmonic mean: 75.83 -> 77.65 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MaPLe — MaPLe: Multi-Modal Prompt Learning (CVPR 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Learning to Prompt with Text Only Supervision for Vision-Language Models (AAAI 2025 · similarity 0.90)
- Prompt Tuning In a Compact Attribute Space (AAAI 2025 · similarity 0.90)
- Aggregate-and-Adapt Natural Language Prompts for Downstream Generalization of CLIP (NeurIPS 2024 · similarity 0.90)
- PromptKD: Unsupervised Prompt Distillation for Vision-Language Models (CVPR 2024 · similarity 0.90)
- IPO: Interpretable Prompt Optimization for Vision-Language Models (NeurIPS 2024 · similarity 0.90)
- Learning to Learn Better Visual Prompts (AAAI 2024 · similarity 0.89)
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.
