Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
TransVLAD: Focusing on Locally Aggregated Descriptors for Few-Shot Learning
Haoquan Li, Laoming Zhang, Daoan Zhang, Lang Fu, Peng Yang, Jianguo Zhang
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 all four components (MAE pre-training, NeXtVLAD, masked fine-tuning, soft focal loss) (ablation) | ViT-B trained directly · miniImageNet 5-way 1-shot | accuracy (%): 34.55 -> 68.24 | Improved |
| add all four components (ablation) | ViT-B trained directly · miniImageNet 5-way 5-shot | accuracy (%): 45.73 -> 84.42 | Improved |
| add masking operation in fine-tuning (ablation) | TransVLAD · miniImageNet 5-way 1-shot | accuracy (%): +2.8% | Improved |
| mask 70% of patches in fine-tuning (ablation) | TransVLAD · training speed | accuracy (%): more than 3 times faster | Improved |
| add NeXtVLAD instead of mean pooling (ablation) | TransVLAD · miniImageNet | accuracy (%) | Improved |
| add MAE pre-training before fine-tuning (ablation) | ViT transformer · miniImageNet | accuracy (%) | Improved |
| replace cross-entropy with soft focal loss (ablation) | TransVLAD · miniImageNet | accuracy (%) | Improved |
| set focusing parameter gamma to 2 (ablation) | TransVLAD · miniImageNet validation | accuracy (%): best at gamma = 2 | Improved |
| set masking ratio to 70% (ablation) | TransVLAD · miniImageNet validation | accuracy (%): best at 70% | Improved |
| train on larger tieredImageNet | TransVLAD vs SSFormers · tieredImageNet 5-way 1-shot | accuracy (%): 72.52 -> 77.20 | Improved |
| transfer from miniImageNet to CUB | TransVLAD vs S2M2 · CUB cross-domain 5-way 1-shot | accuracy (%): 48.42 -> 50.45 | Improved |
| train on CUB and test on other datasets | TransVLAD · CUB->miniImageNet | accuracy (%): 68.24 -> 39.36 | Hurt |
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)
- tSF: Transformer-Based Semantic Filter for Few-Shot Learning (ECCV 2022 · similarity 0.89)
- One Meta-tuned Transformer is What You Need for Few-shot Learning (ICML 2024 · similarity 0.88)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.87)
- Closest Neighbors are Harmful for Lightweight Masked Auto-encoders (CVPR 2025 · similarity 0.86)
- Unlocking the Potential of Pre-trained Vision Transformers for Few-Shot Semantic Segmentation through Relationship Descriptors (CVPR 2024 · similarity 0.86)
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.
