Research map / Detection and segmentation / CNN features
RefineMask: Towards High-Quality Instance Segmentation With Fine-Grained Features
Check 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 the Mask R-CNN mask head with RefineMask multi-stage refinement and fine-grained features | Mask R-CNN, R50-FPN · COCO val2017 | AP: 34.7 -> 37.3 | Helped |
| Replace the Mask R-CNN mask head with RefineMask multi-stage refinement and fine-grained features | Mask R-CNN, R50-FPN · COCO val2017, LVIS annotations | AP*: 36.8 -> 40.9 | Helped |
| Replace the Mask R-CNN mask head with RefineMask | Mask R-CNN, R50-FPN · LVISv1.0 validation | AP: 22.1 -> 25.5 | Helped |
| Add RefineMask multi-stage refinement head | lvisTraveler · LVISv1.0 test-dev | AP: 41.2 -> 42.5 | Helped |
| Replace the Mask R-CNN mask head with RefineMask | Mask R-CNN, R50-FPN · Cityscapes validation | AP: 33.8 -> 37.6 | Helped |
| Add a second refinement stage and increase output size (ablation) | RefineMask · COCO val2017 | AP: 35.7 -> 36.6 | Helped |
| Increase refinement from 2 to 3 stages (ablation) | RefineMask · COCO val2017 | AP: 36.6 -> 37.3 | Helped |
| Increase refinement from 3 to 4 stages (ablation) | RefineMask · COCO val2017 | AP: 37.3 -> 37.1 | Hurt |
| Use SFM instead of a single 1×1 convolution for feature fusion (ablation) | RefineMask mask head · COCO val2017 | AP: 35.8 -> 37.3 | Helped |
| Add the semantic head and fine-grained features (ablation) | RefineMask · COCO val2017 | AP: 35.7 -> 37.3 | Helped |
| Add multi-stage refinement and supervision (ablation) | RefineMask · COCO val2017 | AP*: 39.5 -> 40.9 | Helped |
| Add boundary-aware refinement (ablation) | RefineMask · COCO val2017 | AP*: 40.0 -> 40.9 | Helped |
Closest papers by meaning
- Look Closer To Segment Better: Boundary Patch Refinement for Instance Segmentation (CVPR 2021 · similarity 0.90)
- BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation (CVPR 2020 · similarity 0.90)
- PatchDCT: Patch Refinement for High Quality Instance Segmentation (ICLR 2023 · similarity 0.88)
- Instance Segmentation With Mask-Supervised Polygonal Boundary Transformers (CVPR 2022 · similarity 0.88)
- DynaMask: Dynamic Mask Selection for Instance Segmentation (CVPR 2023 · similarity 0.88)
- Training-Free Dataset Pruning for Instance Segmentation (ICLR 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.