Research map / Detection and segmentation / Weakly supervised detection and segmentation
SIM: Semantic-Aware Instance Mask Generation for Box-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 |
|---|---|---|---|
| replace BoxInst supervision with semantic-aware prototypes and self-correction | Box-supervised instance segmentation · COCO test-dev, 3×, ResNet-101-FPN | AP: 33.2 -> 35.3 | Helped |
| add semantic-aware instance mask generation to BoxInst | Box-supervised instance segmentation · PASCAL VOC val2012, ResNet-50 | AP: 34.3 -> 36.7 | Helped |
| add pseudo mask loss (ablation) | BoxInst baseline · COCO val2017 | AP: 30.7 -> 31.9 | Helped |
| add online weakly-supervised Copy-Paste loss (ablation) | SIM with pseudo mask loss · COCO val2017 | AP: 31.9 -> 32.2 | Helped |
| add self-correction modulation (α=0.5) instead of using semantic masks alone (ablation) | SIM · COCO val2017 | AP: 30.5 -> 32.2 | Helped |
| retain semantic-mask contribution rather than use instance masks alone (α=0.5 instead of α=1) (ablation) | SIM · COCO val2017 | AP: 31.4 -> 32.2 | Helped |
| increase prototypes per category from 1 to 5 (ablation) | SIM · COCO val2017 | AP: 31.6 -> 32.0 | Helped |
| increase prototypes per category from 10 to 50 (ablation) | SIM · COCO val2017 | AP: 32.2 -> 32.2 | No clear effect |
| weight positive samples by box IoU rather than equally (μ=5 instead of μ=0) (ablation) | SIM · COCO val2017 | AP: 31.4 -> 32.2 | Helped |
| set positive-sample weighting μ to 20 instead of 5 (ablation) | SIM · COCO val2017 | AP: 32.2 -> 31.3 | Hurt |
| set low/high confidence thresholds to 0.5/0.5 instead of 0.3/0.7 (ablation) | SIM · COCO val2017 | AP: 32.2 -> 31.4 | Hurt |
| set low/high confidence thresholds to 0.2/0.8 instead of 0.3/0.7 (ablation) | SIM · COCO val2017 | AP: 32.2 -> 31.8 | Hurt |
Closest papers by meaning
- BoxInst: High-Performance Instance Segmentation With Box Annotations (CVPR 2021 · similarity 0.92)
- BoxTeacher: Exploring High-Quality Pseudo Labels for Weakly Supervised Instance Segmentation (CVPR 2023 · similarity 0.91)
- Fully Convolutional Instance-Aware Semantic Segmentation (CVPR 2017 · similarity 0.90)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 2024 · similarity 0.89)
- Extreme Point Supervised Instance Segmentation (CVPR 2024 · similarity 0.88)
- WISH: Weakly Supervised Instance Segmentation using Heterogeneous Labels (CVPR 2025 · similarity 0.88)
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.