Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Universal Few-shot Learning of Dense Prediction Tasks with Visual Token Matching
Donggyun Kim, Jinwoo Kim, Seongwoong Cho, Chong Luo, Seunghoon Hong
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 |
|---|---|---|---|
| use VTM rather than the strongest fully supervised baseline | Taskonomy SS, 10-shot · Taskonomy SS | mIoU: 0.4449 -> 0.4097 | Hurt |
| use VTM rather than the strongest fully supervised baseline | Taskonomy SN, 10-shot · Taskonomy SN | mErr: 6.4414 -> 11.4391 | Hurt |
| use VTM rather than the strongest fully supervised baseline | Taskonomy ED, 10-shot · Taskonomy ED | RMSE: 0.0534 -> 0.0741 | Hurt |
| use VTM rather than the strongest fully supervised baseline | Taskonomy ZD, 10-shot · Taskonomy ZD | RMSE: 0.0268 -> 0.0316 | Hurt |
| use VTM rather than the strongest fully supervised baseline | Taskonomy TE, 10-shot · Taskonomy TE | RMSE: 0.0188 -> 0.0791 | Hurt |
| use VTM rather than the strongest fully supervised baseline | Taskonomy OE, 10-shot · Taskonomy OE | RMSE: 0.0689 -> 0.0912 | Hurt |
| use VTM rather than the strongest fully supervised baseline | Taskonomy K2, 10-shot · Taskonomy K2 | RMSE: 0.0358 -> 0.0639 | Hurt |
| use VTM rather than the strongest fully supervised baseline | Taskonomy K3, 10-shot · Taskonomy K3 | RMSE: 0.0357 -> 0.0519 | Hurt |
| use VTM rather than the strongest fully supervised baseline | Taskonomy RS, 10-shot · Taskonomy RS | RMSE: 0.0860 -> 0.1089 | Hurt |
| use VTM rather than the strongest fully supervised baseline | Taskonomy PC, 10-shot · Taskonomy PC | RMSE: 0.0347 -> 0.0420 | Hurt |
| add task-specific adaptation (ablation) | VTM on Taskonomy SS · Taskonomy SS | mIoU: 0.0002 -> 0.4097 | Improved |
| add non-parametric matching instead of a parametric linear mapping (ablation) | VTM on Taskonomy SS · Taskonomy SS | mIoU: 0.2681 -> 0.4097 | Improved |
Closest papers by meaning
- All Tokens Matter: Token Labeling for Training Better Vision Transformers (NeurIPS 2021 · similarity 0.88)
- Dense Classification and Implanting for Few-Shot Learning (CVPR 2019 · similarity 0.88)
- Visual Prompting for Generalized Few-shot Segmentation: A Multi-scale Approach (CVPR 2024 · similarity 0.87)
- DenseCLIP: Language-Guided Dense Prediction With Context-Aware Prompting (CVPR 2022 · similarity 0.87)
- Unlocking the Potential of Pre-trained Vision Transformers for Few-Shot Semantic Segmentation through Relationship Descriptors (CVPR 2024 · similarity 0.86)
- Pixels to Tokens: Token Space Efficient Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2026 · 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.
