Research map / Detection and segmentation / CNN features
REFINE: Prediction Fusion Network for Panoptic 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 |
|---|---|---|---|
| Add IBPP to the Panoptic-FPN baseline (ablation) | Panoptic-FPN with ResNet-50 · COCO val | PQ: 39.6 -> 42.2 | Helped |
| Add IBPP to the Panoptic-FPN baseline (ablation) | Panoptic-FPN with ResNet-101 · COCO val | PQ: 40.9 -> 43.4 | Helped |
| Increase IBPP refinement from one to two iterations (ablation) | IBPP with ResNet-50 · COCO val | PQ: 41.0 -> 41.6 | Helped |
| Increase IBPP refinement from two to three iterations (ablation) | IBPP with ResNet-50 · COCO val | PQ: 41.6 -> 42.2 | Helped |
| Add soft occlusion estimation with σ=2.5 (ablation) | REFINE occlusion estimator · COCO val | PQth: 58.5 -> 58.9 | Helped |
| Use σ=4.0 rather than σ=2.5 in SOE (ablation) | REFINE occlusion estimator · COCO val | PQth: 58.9 -> 58.6 | Hurt |
| Use σ=2.0 rather than σ=2.5 in SOE (ablation) | REFINE occlusion estimator · COCO val | PQth: 58.9 -> 58.8 | Hurt |
| Add appearance features to the mask-prediction shape prior (ablation) | Pairwise occlusion estimator · COCO val | PQth: 57.4 -> 58.4 | Helped |
| Add class-label prior to shape and appearance priors (ablation) | Pairwise occlusion estimator · COCO val | PQth: 58.4 -> 58.5 | Helped |
| Add confidence prior with soft penalization (ablation) | Pairwise occlusion estimator · COCO val | PQth: 58.5 -> 58.9 | Helped |
| Add Unknown Erasing after IBPP and SOE (ablation) | REFINE with ResNet-50 · COCO val | PQ: 44.2 -> 44.9 | Helped |
| Replace Cascade Panoptic-FPN with REFINE | ResNet-101-DCN panoptic segmentation · COCO val | PQ: 46.1 -> 49.3 | Helped |
Closest papers by meaning
- Unifying Training and Inference for Panoptic Segmentation (CVPR 2020 · similarity 0.89)
- PanopticDepth: A Unified Framework for Depth-Aware Panoptic Segmentation (CVPR 2022 · similarity 0.88)
- Combinatorial Optimization for Panoptic Segmentation: A Fully Differentiable Approach (NeurIPS 2021 · similarity 0.88)
- Fully Convolutional Networks for Panoptic Segmentation (CVPR 2021 · similarity 0.88)
- LPSNet: A Lightweight Solution for Fast Panoptic Segmentation (CVPR 2021 · similarity 0.88)
- Task-aligned Part-aware Panoptic Segmentation through Joint Object-Part Representations (CVPR 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.