Research map / Detection and segmentation / CNN features
InstaBoost: Boosting Instance Segmentation via Probability Map Guided Copy-Pasting
Hao-Shu Fang, Jianhua Sun, Runzhong Wang, Minghao Gou, Yong-Lu Li, Cewu Lu
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 random object jittering (random InstaBoost) | Mask R-CNN Res-50-FPN · COCO test-dev segmentation | AP seg: 33.8 -> 35.5 | Improved |
| add appearance-consistency heatmap guidance on top of random jitter (ablation) | Mask R-CNN Res-50-FPN · COCO test-dev segmentation | AP seg: 35.5 -> 36.0 | Improved |
| add InstaBoost (heatmap guided) | Mask R-CNN Res-101-FPN · COCO test-dev segmentation | AP seg: 35.7 -> 37.9 | Improved |
| add InstaBoost (heatmap guided) | Cascade R-CNN Res-101-FPN · COCO test-dev segmentation | AP seg: 37.3 -> 39.5 | Improved |
| add InstaBoost (heatmap guided) | Mask R-CNN Res-101-FPN · COCO test-dev detection | AP det: 38.2 -> 43.0 | Improved |
| add InstaBoost (heatmap guided) | Mask R-CNN Res-50-FPN · VOCSDS segmentation | AP seg: 38.88 -> 42.73 | Improved |
| replace appearance-guided placement with random paste (ablation) | Mask R-CNN Res-50-FPN · COCO segmentation | AP seg: 33.8 -> 32.7 | Hurt |
| replace appearance-guided placement with random paste (ablation) | Mask R-CNN Res-50-FPN · VOC segmentation | AP seg: 38.88 -> 37.58 | Hurt |
| use InstaBoost instead of a context-model augmentation (ablation) | Mask R-CNN Res-101-FPN · COCO segmentation | AP seg: 36.2 -> 37.9 | Improved |
| increase InstaBoost training cost (train speed) | Mask R-CNN Res-101-FPN · COCO | train speed s/iter: 1.68 -> 1.71 | No clear effect |
| set translation ratio to 1/15 instead of 1 (ablation) | InstaBoost · COCO val segmentation | AP seg: 34.51 -> 35.10 | Improved |
| narrow scaling range from 0.7-1.3 to 0.8-1.2 (ablation) | InstaBoost · COCO val segmentation | AP seg: 34.85 -> 35.10 | Improved |
Closest papers by meaning
- Training-Free Dataset Pruning for Instance Segmentation (ICLR 2025 · similarity 0.88)
- MaskBooster: End-to-End Self-Training for Sparsely Supervised Instance Segmentation (AAAI 2023 · similarity 0.88)
- YOLACT: Real-Time Instance Segmentation (ICCV 2019 · similarity 0.88)
- SOLO: Segmenting Objects by Locations (ECCV 2020 · similarity 0.88)
- Sparse Instance Activation for Real-Time Instance Segmentation (CVPR 2022 · similarity 0.87)
- Learning to Segment Every Thing (CVPR 2018 · 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.
