Research map / Vision-language and multimodal / CLIP and zero-shot
LASP: Text-to-Text Optimization for Language-Aware Soft Prompting of Vision & Language Models
Adrian Bulat, Georgios Tzimiropoulos
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 the text-to-text cross-entropy loss to CoOp (ablation) | CoOp soft prompting · 11-dataset average, new classes | accuracy: 63.22 -> 71.54 | Improved |
| add grouped prompts to Text-to-Text (ablation) | LASP · 11-dataset average, new classes | accuracy: 71.54 -> 73.48 | Improved |
| add visual/text re-alignment (ablation) | Grouped LASP · 11-dataset average, new classes | accuracy: 73.48 -> 74.90 | Improved |
| add virtual class names during training (ablation) | LASP · 11-dataset average, new classes | accuracy: 74.90 -> 76.11 | Improved |
| increase hand-crafted template count from 34 to 100 (ablation) | Text-to-Text prompting · DTD new classes | accuracy: 52.10 -> 56.53 | Improved |
| replace hand-crafted templates with random templates (ablation) | Text-to-Text prompting · DTD new classes | accuracy: 52.10 -> 51.63 | Hurt |
| replace text-to-text cross-entropy with L1 loss (ablation) | LASP Text-to-Text objective · new classes | accuracy: 71.54 -> 66.01 | Hurt |
| replace text-to-text cross-entropy with L2 loss (ablation) | LASP Text-to-Text objective · new classes | accuracy: 71.54 -> 65.80 | Hurt |
| train with virtual classes (ablation) | LASP · EuroSAT new classes | accuracy: 77.78 -> 83.37 | Improved |
| train with virtual classes (ablation) | LASP · OxfordPets new classes | accuracy: 97.93 -> 97.87 | No clear effect |
| train with virtual classes (ablation) | LASP · Flowers102 new classes | accuracy: 74.0 -> 73.53 | Hurt |
| intermix template categories during training (ablation) | LASP · novel classes | accuracy: 0.5% | Hurt |
Later papers that compared with this method
3 later accepted papers report LASP as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Knowledge-Enhanced Explainable Prompting for Vision-Language Models (AAAI 2026 · better on 5 of 5 shared results)
- FATE: Feature-Adapted Parameter Tuning for Vision-Language Models (AAAI 2025 · better on 1 of 2 shared results)
- ArGue: Attribute-Guided Prompt Tuning for Vision-Language Models (CVPR 2024 · better on 12 of 12 shared results)
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
- LaViP: Language-Grounded Visual Prompting (AAAI 2024 · similarity 0.90)
- Make Prompts Adaptable: Bayesian Modeling for Vision-Language Prompt Learning with Data-Dependent Prior (AAAI 2024 · similarity 0.90)
- Learning to Learn Better Visual Prompts (AAAI 2024 · similarity 0.89)
- Learning to Prompt with Text Only Supervision for Vision-Language Models (AAAI 2025 · similarity 0.89)
- ProAPO: Progressively Automatic Prompt Optimization for Visual Classification (CVPR 2025 · similarity 0.89)
- PromptKD: Unsupervised Prompt Distillation for Vision-Language Models (CVPR 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.
