Research map / Detection and segmentation / Weakly supervised detection and segmentation
BoxTeacher: Exploring High-Quality Pseudo Labels for Weakly Supervised Instance Segmentation
Read the paperCheck 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 |
|---|---|---|---|
| Train with BoxTeacher pseudo-mask losses rather than box-supervised loss alone (ablation) | BoxTeacher on COCO · COCO | mask AP: 30.7 -> 32.2 | Helped |
| Use dice rather than BCE for pseudo-mask pixel loss (ablation) | BoxTeacher · COCO | mask AP: 28.9 -> 31.8 | Helped |
| Add affinity loss to dice pixel loss (ablation) | BoxTeacher · COCO | mask AP: +0.4 AP | Helped |
| Set pixel-loss weight to 0.1 instead of 0.5 (ablation) | BoxTeacher · COCO | mask AP: 31.8 -> 31.4 | Hurt |
| Set pixel-loss weight to 1.0 instead of 0.5 (ablation) | BoxTeacher · COCO | mask AP: 31.8 -> 31.5 | Hurt |
| Set affinity-loss weight to 0.5 instead of 0.1 (ablation) | BoxTeacher · COCO | mask AP: 32.2 -> 31.7 | Hurt |
| Use mask-aware confidence instead of classification, IoU, or mean-entropy scores (ablation) | BoxTeacher · COCO | mask AP: 32.0, 32.2, 31.8 -> 32.6 | Helped |
| Use naive self-training with BoxInst pseudo masks instead of end-to-end BoxTeacher | CondInst / BoxTeacher · COCO | mask AP: 31.3 -> 34.2 | Hurt |
| Add strong augmentation at 1× schedule (ablation) | BoxTeacher · COCO | mask AP: 32.6 -> 32.2 | Hurt |
| Add strong augmentation at 3× schedule (ablation) | BoxTeacher · COCO | mask AP: +0.6 AP | Helped |
| Add strong augmentation at 3× schedule (ablation) | CondInst · COCO | mask AP: +0.1 AP | No clear effect |
| Add EMA (ablation) | BoxInst · COCO | mask AP: 30.6 -> 30.7 | No clear effect |
Closest papers by meaning
- BoxInst: High-Performance Instance Segmentation With Box Annotations (CVPR 2021 · similarity 0.93)
- SIM: Semantic-Aware Instance Mask Generation for Box-Supervised Instance Segmentation (CVPR 2023 · similarity 0.91)
- The Devil Is in the Points: Weakly Semi-Supervised Instance Segmentation via Point-Guided Mask Representation (CVPR 2023 · similarity 0.91)
- Extreme Point Supervised Instance Segmentation (CVPR 2024 · similarity 0.91)
- WISH: Weakly Supervised Instance Segmentation using Heterogeneous Labels (CVPR 2025 · similarity 0.89)
- Sketchy Bounding-box Supervision for 3D Instance Segmentation (CVPR 2025 · 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.