Research map / Detection and segmentation / Weakly supervised detection and segmentation
Fully Convolutional Networks for Panoptic Segmentation with Point-based Supervision
Yanwei Li, Hengshuang Zhao, Xiaojuan Qi, Yukang Chen, Lu Qi, Liwei Wang, Zeming Li, Jian Sun, Jiaya Jia
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 kernel-level fusion (ablation) | Panoptic FCN · COCO val | PQ (%): 38.7 -> 41.3 | Improved |
| add Matrix NMS on top of kernel fusion (ablation) | Panoptic FCN · COCO val | PQ (%): 41.3 -> 41.3 | No clear effect |
| add relative coordinates to both kernel head and feature encoder (ablation) | Panoptic FCN · COCO val | PQ (%): 39.9 -> 41.3 | Improved |
| add coordinates to kernel head only (ablation) | Panoptic FCN · COCO val | PQ (%): 39.9 -> 39.9 | No clear effect |
| add deformable convolutions to position head (ablation) | Panoptic FCN · COCO val | PQ (%): 39.2 -> 39.9 | Improved |
| add weighted dice loss sampling k=7 kernels (ablation) | Panoptic FCN · COCO val | PQ (%): 40.2 -> 41.3 | Improved |
| use mass center instead of box center for ground truth (ablation) | Panoptic FCN · COCO val | PQ (%): 40.6 -> 41.3 | Improved |
| use Semantic FPN feature instead of FPN-P2 (ablation) | Panoptic FCN · COCO val | PQ (%): 40.6 -> 41.3 | Improved |
| raise feature encoder channels from 16 to 64 (ablation) | Panoptic FCN · COCO val | PQ (%): 39.9 -> 41.3 | Improved |
| add dilated shape augmentation to P20 supervision (ablation) | Panoptic FCN P20 · COCO val | PQ (%): 32.7 -> 34.0 | Improved |
| increase training schedule from 1x to 2x (ablation) | Panoptic FCN · COCO val | PQ (%): 41.3 -> 43.2 | Improved |
| use concave instead of convex target with P100 points (ablation) | Panoptic FCN P100 · COCO val | PQ (%): 33.4 -> 37.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- SOLO V2 — SOLOv2: Dynamic and Fast Instance Segmentation (NeurIPS 2020) · better on 1 of 1 shared result
- UPSNet — UPSNet: A Unified Panoptic Segmentation Network (CVPR 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
- Fully Convolutional Networks for Panoptic Segmentation (CVPR 2021 · similarity 0.96)
- Unifying Training and Inference for Panoptic Segmentation (CVPR 2020 · similarity 0.91)
- Weakly- and Semi-Supervised Panoptic Segmentation (ECCV 2018 · similarity 0.91)
- Pointly-Supervised Panoptic Segmentation (ECCV 2022 · similarity 0.91)
- Scene-Centric Unsupervised Panoptic Segmentation (CVPR 2025 · similarity 0.90)
- Fully Data-Driven Pseudo Label Estimation for Pointly-Supervised Panoptic Segmentation (AAAI 2024 · similarity 0.88)
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.
