Research map / Detection and segmentation / CNN features
Panoptic Feature Pyramid Networks
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 |
|---|---|---|---|
| Use one ResNet-101-FPN for both tasks rather than two separate ResNet-50-FPN networks | Panoptic FPN · COCO panoptic segmentation | PQ: 39.2 -> 40.3 (+1.1) | Helped |
| Use one ResNet-101-FPN for both tasks rather than two separate ResNet-50-FPN networks | Panoptic FPN · Cityscapes panoptic segmentation | PQ: 57.7 -> 58.1 (+0.4) | Helped |
| Add semantic segmentation loss with tuned weighting | Mask R-CNN on COCO · COCO instance segmentation | AP: 33.9 -> 34.0 | Helped |
| Add instance segmentation loss with tuned weighting | Semantic FPN on COCO · COCO semantic segmentation | mIoU: 40.2 -> 41.5 | Helped |
| Use 128 semantic-branch channels instead of 64 (ablation) | Semantic FPN · COCO semantic segmentation | mIoU: 39.6 -> 40.2 | Helped |
| Increase semantic-branch width from 128 to 256 channels (ablation) | Semantic FPN · COCO semantic segmentation | mIoU: 40.2 -> 40.1 | No clear effect |
| Use 128 semantic-branch channels instead of 64 (ablation) | Semantic FPN · Cityscapes semantic segmentation | mIoU: 74.1 -> 74.5 | Helped |
| Replace feature summation with concatenation (ablation) | Semantic FPN · COCO semantic segmentation | mIoU: 40.2 -> 39.9 | Hurt |
| Replace feature summation with concatenation (ablation) | Semantic FPN · Cityscapes semantic segmentation | mIoU: 74.5 -> 74.4 | No clear effect |
| Compute instance and semantic losses together each minibatch instead of alternating them (ablation) | Panoptic FPN · COCO panoptic segmentation | PQ: 37.5 -> 39.0 (+1.5) | Helped |
| Compute instance and semantic losses together each minibatch instead of alternating them (ablation) | Panoptic FPN · Cityscapes panoptic segmentation | PQ: 57.4 -> 57.7 (+0.3) | Helped |
| Split FPN channels into separate task-specific groups (ablation) | Panoptic FPN · COCO panoptic segmentation | PQ: 39.0 -> 38.8 | Hurt |
Closest papers by meaning
- K-Net: Towards Unified Image Segmentation (NeurIPS 2021 · similarity 0.90)
- Unifying Training and Inference for Panoptic Segmentation (CVPR 2020 · similarity 0.90)
- Instance-Aware Semantic Segmentation via Multi-Task Network Cascades (CVPR 2016 · similarity 0.90)
- Fully Convolutional Networks for Panoptic Segmentation (CVPR 2021 · similarity 0.90)
- LPSNet: A Lightweight Solution for Fast Panoptic Segmentation (CVPR 2021 · similarity 0.90)
- Seamless Scene Segmentation (CVPR 2019 · 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.