Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Learning What Not To Segment: A New Perspective on Few-Shot Segmentation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 a base-class learner and ensemble its predictions with the meta learner | BAM · PASCAL-5ᵢ, 1-shot, ResNet50 | mIoU: 65.40 -> 67.81 | Helped |
| Add a base-class learner and ensemble its predictions with the meta learner | BAM · COCO-20ᵢ, 1-shot, ResNet50 | mIoU: 43.09 -> 46.23 | Helped |
| Pre-train the base learner (ablation) | PASCAL-5ᵢ 1-shot model · PASCAL-5ᵢ, 1-shot, VGG16 | mIoU: 57.61 -> 59.12 | Helped |
| Add the meta learner loss (ablation) | PASCAL-5ᵢ 1-shot model · PASCAL-5ᵢ, 1-shot, VGG16 | mIoU: 59.12 -> 59.76 | Helped |
| Initialize ensemble weights specifically rather than randomly (ablation) | BAM ensemble · PASCAL-5ᵢ, 1-shot, VGG16 | mIoU: 2.73% mIoU improvements | Helped |
| Add the scene-difference adjustment factor ψ (ablation) | BAM ensemble · PASCAL-5ᵢ, 1-shot, VGG16 | mIoU: 62.49 -> 64.41 | Helped |
| Use adaptive support reweighting instead of alternative 5-shot fusion schemes (ablation) | BAM · PASCAL-5ᵢ, 5-shot, VGG16 | mIoU: 64.41 -> 68.76; Mask-OR 65.15; Mask-Avg 65.92; Feature-Avg 66.83 | Helped |
| Replace pixel-wise support labels with bounding boxes (ablation) | BAM · PASCAL-5ᵢ, 1-shot, VGG16 | mIoU: 64.41 -> 62.25 | Hurt |
| Replace pixel-wise support labels with bounding boxes (ablation) | BAM · PASCAL-5ᵢ, 5-shot, VGG16 | mIoU: 68.76 -> 66.17 | Hurt |
| Add the ensemble module (ablation) | BAM generalized FSS · PASCAL-5ᵢ generalized FSS, 1-shot, VGG16 | novel-class mIoU: 37.54 -> 43.19 | Helped |
| Add the ensemble module (ablation) | BAM generalized FSS · PASCAL-5ᵢ generalized FSS, 1-shot, VGG16 | all-class mIoU: 59.65 -> 61.07 | Helped |
| Train the base and meta learners separately rather than jointly (ablation) | BAM · PASCAL-5ᵢ, 1-shot, VGG16 | mIoU | Helped |
Closest papers by meaning
- Adaptive FSS: A Novel Few-Shot Segmentation Framework via Prototype Enhancement (AAAI 2024 · similarity 0.92)
- A Surprisingly Simple Approach to Generalized Few-Shot Semantic Segmentation (NeurIPS 2024 · similarity 0.91)
- Label-Efficient Few-Shot Semantic Segmentation with Unsupervised Meta-Training (AAAI 2024 · similarity 0.91)
- Unlocking the Potential of Pre-trained Vision Transformers for Few-Shot Semantic Segmentation through Relationship Descriptors (CVPR 2024 · similarity 0.90)
- Adapt Before Comparison: A New Perspective on Cross-Domain Few-Shot Segmentation (CVPR 2024 · similarity 0.90)
- Bridge the Points: Graph-based Few-shot Segment Anything Semantically (NeurIPS 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.