Research map / Detection and segmentation / CNN features
IMP: Instance Mask Projection for High Accuracy Semantic Segmentation of Things
Cheng-Yang Fu, Tamara L. Berg, Alexander C. Berg
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 Instance Mask Projection (ablation) | Panoptic-P2 · Varied Clothing (val) | semantic mIOU: 37.14 -> 46.59 | Improved |
| add Instance Mask Projection (ablation) | Panoptic-FPN · Varied Clothing (val) | semantic mIOU: 45.01 -> 47.03 | Improved |
| add Instance Mask Projection (ablation) | Panoptic-FPN · Varied Clothing (val) | semantic mAcc: 57.08 -> 61.52 | Improved |
| add Instance Mask Projection (ablation) | Panoptic-P2 · ModaNet (val) | semantic mIOU: 65.93 -> 69.65 | Improved |
| add Instance Mask Projection without augmentation (ablation) | Panoptic-FPN · Cityscapes val | mIOU: 72.74 -> 74.09 | Improved |
| add Instance Mask Projection with all augmentations (ablation) | Panoptic-FPN · Cityscapes val | mIOU: 76.11 -> 77.49 | Improved |
| add multi-scale training (ablation) | Panoptic-FPN with ColorJitter · Cityscapes val | mIOU: 73.12 -> 75.99 | Improved |
| add hard bootstrapping with ratio 0.10 (ablation) | Panoptic-FPN with ColorJitter · Cityscapes val | mIOU: 73.12 -> 74.94 | Improved |
| add color jitter (ablation) | Panoptic-FPN · Cityscapes val | mIOU: 72.74 -> 73.12 | No clear effect |
| use IMP-projected Mask R-CNN masks directly instead of a semantic head (ablation) | Mask R-CNN-IMP · Varied Clothing (val) | semantic mIOU: 42.66 -> 43.91 | Improved |
| add IMP to stuff classes in Cityscapes (wall) (ablation) | Panoptic-FPN · Cityscapes val | per-class IOU: -1.6/-6.7 | Hurt |
| add IMP for Cityscapes thing classes (ablation) | Panoptic-FPN · Cityscapes val (thing classes) | mIOU: +4.2/+3.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PSANet101 — PSANet: Point-wise Spatial Attention Network for Scene Parsing (ECCV 2018) · 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
- Boundary-Aware Instance Segmentation (CVPR 2017 · similarity 0.87)
- Mask Encoding for Single Shot Instance Segmentation (CVPR 2020 · similarity 0.86)
- Segment Anything, Even Occluded (CVPR 2025 · similarity 0.86)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 2024 · similarity 0.85)
- SegFace: Face Segmentation of Long-Tail Classes (AAAI 2025 · similarity 0.85)
- MaskDiff: Modeling Mask Distribution with Diffusion Probabilistic Model for Few-Shot Instance Segmentation (AAAI 2024 · similarity 0.85)
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.
