Research map / Detection and segmentation / Open-set and few-shot detection
Scaling Up Instance Annotation via Label Propagation
Dim P. Papadopoulos, Ethan Weber, Antonio Torralba
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 |
|---|---|---|---|
| cluster on mask attention features combined with mask score instead of other feature types (ablation) | hierarchical clustering · ADE20K simulated | annotations vs clusters | Improved |
| use heuristics-only search with mask score instead of BFS or DFS or universal thresholding (ablation) | tree search · ADE20K simulated | annotations vs clusters | Improved |
| set K_pa high (very high values) when selecting clusters (ablation) | active cluster selection · ADE20K simulated | number of good annotations | Hurt |
| use low K_pa values rather than K_pa=0.7 (ablation) | active cluster selection · ADE20K simulated | annotation performance | No clear effect |
| increase initial training set size from 500 to 10,000 images (ablation) | label propagation · ADE20K simulated | annotation quantity | Improved |
| use label propagation pipeline instead of confidence-score filtering | PointRend predictions · ADE20K simulated | annotations at matched quality: 67 -> 15,628 | Improved |
| verify all masks per cluster (N_s = infinity) instead of N_s = 15 (ablation) | label propagation · Large-scale simulation | SQ: 0.85 -> 0.89 | Improved |
| train detector on propagated labels plus initial set instead of initial set only | PointRend · ADE20K val | AP: 8.1 -> 12.7 | Improved |
| use propagated labels instead of spending same 73 h on manual annotation | PointRend · ADE20K val | AP: 9.6 -> 12.7 | Improved |
| apply hyperparameters tuned on ADE20K to COCO and OpenImages without retuning | label propagation · COCO and OpenImages | annotation gain | No clear effect |
Closest papers by meaning
- Large-Scale Interactive Object Segmentation With Human Annotators (CVPR 2019 · similarity 0.93)
- Learning to Segment Every Thing (CVPR 2018 · similarity 0.91)
- Object Affinity Learning: Towards Annotation-Free Instance Segmentation (TPAMI 2023 · similarity 0.89)
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · similarity 0.89)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 2024 · similarity 0.87)
- Extreme Point Supervised Instance Segmentation (CVPR 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.
