Research map / Detection and segmentation / CNN features
SOLO: A Simple Framework for Instance Segmentation
Xinlong Wang, Rufeng Zhang, Chunhua Shen, Tao Kong, Lei Li
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 FPN multi-level prediction (pyramid) instead of single grid 24 (ablation) | SOLO · COCO val2017 | mask AP: 29.0 -> 35.8 | Improved |
| add one CoordConv layer (ablation) | SOLO · COCO val2017 | mask AP: 32.2 -> 35.8 | Improved |
| add more CoordConv layers (2 or 3 instead of 1) (ablation) | SOLO · COCO val2017 | mask AP: 35.8 -> 35.7 | No clear effect |
| use Dice loss instead of focal loss (ablation) | SOLO · COCO val2017 | mask AP: 31.6 -> 35.8 | Improved |
| use Dice loss instead of BCE (ablation) | SOLO · COCO val2017 | mask AP: 30.0 -> 35.8 | Improved |
| replace BCE with Dice loss in Mask R-CNN | Mask R-CNN · COCO | mask AP: -0.9 | Hurt |
| deepen head from 4 to 7 convs (ablation) | SOLO · COCO val2017 | mask AP: 34.6 -> 35.8 | Improved |
| change category alignment method (interpolation, adaptive-pool, region-grid) (ablation) | SOLO · COCO | mask AP: +/-0.1% AP | No clear effect |
| use decoupled SOLO instead of vanilla | SOLO R50 · COCO test-dev | mask AP: 36.8 -> 37.4 | Improved |
| add explicit coordinates to kernel and feature branches (ablation) | SOLOv2 R50 · COCO val2017 | mask AP: 36.3 -> 37.8 | Improved |
| use unified mask feature instead of separate (ablation) | SOLOv2 R50 · COCO val2017 | mask AP: 37.3 -> 37.8 | Improved |
| use dynamic head instead of decoupled head (ablation) | SOLOv2 · COCO val2017 | mask AP: 36.6 -> 37.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- BlendMask — BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation (CVPR 2020) · better on 1 of 1 shared result
- PolarMask — PolarMask: Single Shot Instance Segmentation With Polar Representation (CVPR 2020) · better on 1 of 1 shared result
- TensorMask — TensorMask: A Foundation for Dense Object Segmentation (ICCV 2019) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SOLO: Segmenting Objects by Locations (ECCV 2020 · similarity 0.97)
- SOLOv2: Dynamic and Fast Instance Segmentation (NeurIPS 2020 · similarity 0.94)
- Pixelwise Instance Segmentation With a Dynamically Instantiated Network (CVPR 2017 · similarity 0.92)
- InstanceCut: From Edges to Instances With MultiCut (CVPR 2017 · similarity 0.91)
- Open-World Instance Segmentation: Exploiting Pseudo Ground Truth From Learned Pairwise Affinity (CVPR 2022 · similarity 0.91)
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · 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.
