Research map / Detection and segmentation / CNN features
SOLOv2: Dynamic and Fast 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 |
|---|---|---|---|
| dynamic mask kernels + unified mask feature + Matrix NMS (SOLOv2 vs SOLO) | SOLO · COCO test-dev | mask AP: 37.8 -> 39.7 | Helped |
| add explicit coordinates to kernel and feature branches (ablation) | SOLOv2 · COCO val2017 | mask AP: 36.3 -> 37.8 | Helped |
| add coordinates to kernel branch only (ablation) | SOLOv2 · COCO val2017 | mask AP: 36.3 -> 36.3 | No clear effect |
| Matrix NMS instead of Fast NMS (ablation) | SOLOv2 · COCO val2017 | mask AP: 36.2 -> 36.6 | Helped |
| Matrix NMS instead of Soft-NMS (ablation) | SOLOv2 · COCO val2017 | NMS time (ms): 22 -> <1 | Helped |
| unified mask feature instead of separate per-level (ablation) | SOLOv2 · COCO val2017 | mask AP: 37.3 -> 37.8 | Helped |
| kernel input channels 128 -> 256 (ablation) | SOLOv2 · COCO val2017 | mask AP: 37.4 -> 37.8 | Helped |
| kernel 3x3 instead of 1x1 (64 ch) (ablation) | SOLOv2 · COCO val2017 | mask AP: 37.4 -> 37.4 | No clear effect |
| kernel channels 256 -> 512 (ablation) | SOLOv2 · COCO val2017 | mask AP: 37.8 -> 37.7 | No clear effect |
| 3x multi-scale schedule vs 1x (ablation) | SOLOv2 · COCO val2017 | mask AP: 34.8 -> 37.8 | Helped |
Closest papers by meaning
- Mask Encoding for Single Shot Instance Segmentation (CVPR 2020 · similarity 0.90)
- BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation (CVPR 2020 · similarity 0.89)
- PolarMask: Single Shot Instance Segmentation With Polar Representation (CVPR 2020 · similarity 0.89)
- MaskDiff: Modeling Mask Distribution with Diffusion Probabilistic Model for Few-Shot Instance Segmentation (AAAI 2024 · similarity 0.88)
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · similarity 0.87)
- Any3DIS: Class-Agnostic 3D Instance Segmentation by 2D Mask Tracking (CVPR 2025 · 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.