Research map / Detection and segmentation / Open-set and few-shot detection
tSF: Transformer-Based Semantic Filter for Few-Shot Learning
Jinxiang Lai, Siqian Yang, Wenlong Liu, Yi Zeng, Zhongyi Huang, Wenlong Wu, Jun Liu, Bin-Bin Gao, Chengjie Wang
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 tSF neck to PatchProto | PatchProto (ResNet-12) · miniImageNet 5-way 1-shot | accuracy: 68.46 -> 69.74 | Improved |
| add tSF neck to PatchProto | PatchProto (WRN-28) · tieredImageNet 5-way 5-shot | accuracy: 86.85 -> 88.05 | Improved |
| add tSF neck to baseline without neck (ablation) | PatchProto (ResNet-12, no rotation) · miniImageNet 5-way 1-shot | accuracy: 67.47 -> 68.37 | Improved |
| add standard self-attention transformer neck (ablation) | PatchProto (ResNet-12, no neck) · miniImageNet 5-way 1-shot | accuracy: 67.47 -> 63.25 | Hurt |
| add DETR-style decoder neck (ablation) | PatchProto (ResNet-12, no neck) · miniImageNet 5-way 1-shot | accuracy: 67.47 -> 63.55 | Hurt |
| use V=theta instead of K=V=theta (tSF-V variant) (ablation) | tSF · miniImageNet 5-way 1-shot | accuracy: 68.37 -> 61.84 | Hurt |
| use K=V=theta rather than only K=theta (tSF-K variant) (ablation) | tSF · miniImageNet 5-way 1-shot | accuracy: 64.73 -> 68.37 | Improved |
| add learnable projections W_Q, W_K, W_V to tSF (ablation) | tSF · miniImageNet 5-way 1-shot | accuracy: 68.37 -> 68.27 | No clear effect |
| increase tSF attention heads from 1 to 4 (ablation) | tSF · miniImageNet 5-way 1-shot | accuracy: 68.37 -> 68.60 | No clear effect |
| vary dimension n of theta from 1 to 1024 (ablation) | tSF · miniImageNet 5-way (1-shot and 5-shot) | accuracy: within 0.5% | No clear effect |
| add global classification auxiliary task (ablation) | PatchProto+tSF (ResNet-12) · miniImageNet 5-way 1-shot | accuracy: 62.76 -> 68.60 | Improved |
| add rotation classification auxiliary task (ablation) | PatchProto+tSF (ResNet-12) · miniImageNet 5-way 1-shot | accuracy: 62.76 -> 65.57 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DANet — DANet: Divergent Activation for Weakly Supervised Object Localization (ICCV 2019) · better on 2 of 2 shared results
- DeFRCN — DeFRCN: Decoupled Faster R-CNN for Few-Shot Object Detection (ICCV 2021) · 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
- Pushing the Limits of Simple Pipelines for Few-Shot Learning: External Data and Fine-Tuning Make a Difference (CVPR 2022 · similarity 0.90)
- TransVLAD: Focusing on Locally Aggregated Descriptors for Few-Shot Learning (ECCV 2022 · similarity 0.89)
- Unlocking the Potential of Pre-trained Vision Transformers for Few-Shot Semantic Segmentation through Relationship Descriptors (CVPR 2024 · similarity 0.88)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.88)
- Simple Semantic-Aided Few-Shot Learning (CVPR 2024 · similarity 0.87)
- One Meta-tuned Transformer is What You Need for Few-shot Learning (ICML 2024 · similarity 0.87)
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.
