Research map / Vision-language and multimodal / Human and embodied modeling
Semantic Prompt for Few-Shot Image Recognition
Wentao Chen, Chenyang Si, Zhang Zhang, Liang Wang, Zilei Wang, Tieniu Tan
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 semantic prompts with spatial and channel interactions | Visformer-T few-shot classifier · miniImageNet 5-way 1-shot | accuracy: 65.16±0.44 -> 72.31±0.40 | Improved |
| Add semantic prompts with spatial and channel interactions | Visformer-T few-shot classifier · tieredImageNet 5-way 1-shot | accuracy: 72.38±0.50 -> 78.03±0.46 | Improved |
| Add semantic prompts with spatial and channel interactions | Visformer-T few-shot classifier · CIFAR-FS 5-way 1-shot | accuracy: 71.99±0.47 -> 82.18±0.40 | Improved |
| Add semantic prompts with spatial and channel interactions | Visformer-T few-shot classifier · FC100 5-way 1-shot | accuracy: 43.77±0.39 -> 48.53±0.38 | Improved |
| Add RandAug and RepeatAug (ablation) | Visformer-T pre-training · four few-shot datasets, 1-shot | accuracy: +2.45% on average | Improved |
| Add spatial interaction (ablation) | Semantic Prompt · four few-shot datasets, 1-shot | accuracy: 5.89% average improvement | Improved |
| Add channel interaction (ablation) | Semantic Prompt · four few-shot datasets, 1-shot | accuracy: 5.43% average improvement | Improved |
| Combine spatial and channel interactions (ablation) | Semantic Prompt · all four ablation datasets, 1-shot | accuracy: 71.59 -> 72.31; 76.20 -> 78.03; 81.19 -> 82.18; 47.83 -> 48.53 | Improved |
| Use CLIP rather than SBERT or GloVe text embeddings | Semantic Prompt · miniImageNet and tieredImageNet, 1-shot | accuracy: 70.70 -> 72.31; 73.31 -> 78.03 | Mixed |
| Use CLIP rather than SBERT or GloVe text embeddings | Semantic Prompt · miniImageNet and tieredImageNet, 5-shot | accuracy: 83.55 -> 83.42; 88.56 -> 88.55 | Mixed |
| Insert prompts in higher rather than lower feature-extractor layers (ablation) | Semantic Prompt · miniImageNet and CIFAR-FS validation | accuracy | Mixed |
| Replace the linear semantic projector with an MLP (ablation) | Semantic Prompt · miniImageNet 5-way | accuracy: 72.31 -> 72.70; 83.42 -> 83.56 | Improved |
Closest papers by meaning
- Self-Prompt Mechanism for Few-Shot Image Recognition (AAAI 2024 · similarity 0.85)
- Learning to Prompt with Text Only Supervision for Vision-Language Models (AAAI 2025 · similarity 0.84)
- CoPL: Contextual Prompt Learning for Vision-Language Understanding (AAAI 2024 · similarity 0.83)
- Visual and Semantic Prompt Collaboration for Generalized Zero-Shot Learning (CVPR 2025 · similarity 0.83)
- SEC-Prompt:SEmantic Complementary Prompting for Few-Shot Class-Incremental Learning (CVPR 2025 · similarity 0.83)
- ViLT-CLIP: Video and Language Tuning CLIP with Multimodal Prompt Learning and Scenario-Guided Optimization (AAAI 2024 · similarity 0.83)
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.
