Research map / Detection and segmentation / Weakly supervised detection and segmentation
MetaAT: Active Testing for Label-Efficient Evaluation of Dense Recognition Tasks
Sanbao Su, Xin Li, Thang Doan, Sima Behpour, Wenbin He, Liang Gou, Fei Miao, Liu Ren
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 region-level loss to meta model training (ablation) | MetaAT · Pascal VOC UNet image level | ER (%): 10.95 -> 6.21 | Improved |
| add equal-width binning (regression to classification) (ablation) | MetaAT · Pascal VOC UNet image level | ER (%): 8.02 -> 6.21 | Improved |
| use ViT meta model instead of ResNet or MLP (ablation) | MetaAT · Pascal VOC UNet image level | ER (%): 7.69 (ResNet) / 10.73 (MLP) -> 6.21 | Improved |
| add output entropy as meta model input (ablation) | MetaAT · Segmentation region-16 level | ER (%): 10.36 -> 4.27 | Improved |
| add target model output as meta model input (ablation) | MetaAT · Segmentation region-16 level | ER (%): 5.38 -> 4.27 | Improved |
| add original image as meta model input (ablation) | MetaAT · Segmentation region-16 level | ER (%): 4.54 -> 4.27 | No clear effect |
| region-level instead of image-level annotation | MetaAT · Pascal VOC segmentation | labels needed for ER below 10%: 2.5% -> less than 0.05% (region-32), 0.02% (region-16) | Improved |
| MetaAT vs RS on ADE20K FCN image level | MetaAT · ADE20K FCN image level | ER (%): 7.85 (RS) vs 8.08 | No clear effect |
| MetaAT vs random sampling on COCO detection | MetaAT · COCO detection image level | ER (%): 2.19 -> 0.67 | Improved |
Closest papers by meaning
- Prediction-Powered Active Testing (NeurIPS 2026 · similarity 0.84)
- Pixels to Tokens: Token Space Efficient Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2026 · similarity 0.84)
- Plug and Play Active Learning for Object Detection (CVPR 2024 · similarity 0.83)
- Doomed to Re-Annotate, Forever: The ImageNet Story (NeurIPS 2026 · similarity 0.83)
- Memory Efficient Transformer Adapter for Dense Predictions (ICLR 2025 · similarity 0.83)
- Context-Aware Meta-Learning (ICLR 2024 · similarity 0.83)
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.
