Research map / Detection and segmentation / Weakly supervised detection and segmentation
Weakly- and Semi-Supervised Panoptic Segmentation
Qizhu Li, Anurag Arnab, Philip H.S. Torr
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 |
|---|---|---|---|
| train the semantic subnetwork and instance subnetwork end-to-end (weak boxes, with COCO) (ablation) | weakly supervised panoptic model · Pascal VOC val | AP^r_vol: 53.3 -> 55.5 | Improved |
| train end-to-end (weak boxes, without COCO pretraining) (ablation) | weakly supervised panoptic model · Pascal VOC val | AP^r_vol: increase by 3.9 from 51.7 | Improved |
| add iterative refinement of approximate ground truth (ablation) | weakly supervised model · Pascal VOC val | AP^r_vol | No clear effect |
| add iterative refinement of approximate ground truth (ablation) | weakly supervised model · Cityscapes val | IoU and AP^r_vol | Improved |
| use COCO pretraining for the semantic subnetwork (ablation) | weakly supervised model · Pascal VOC val | IoU (weak): 74.3 -> 75.7 | Improved |
| switch COCO annotations from weak to full while VOC stays weak (ablation) | weakly supervised model · Pascal VOC val | AP^r_vol: 55.5 -> 56.1 | No clear effect |
| switch VOC annotations from weak to full while COCO stays weak (ablation) | weakly supervised model · Pascal VOC val | AP^r_vol: 55.5 -> 58.9 | Improved |
| switch from weak to full supervision | semantic segmentation model · Pascal VOC val | IoU: 74.3 -> 77.3 | Improved |
| rank stuff instances by mean segmentation confidence instead of constant score (ablation) | weakly supervised model · Cityscapes val (stuff classes) | AP^r_vol: 26.7 -> 33.1 | Improved |
| rank thing instances by mean segmentation confidence instead of detection score (ablation) | weakly supervised model · Cityscapes val (thing classes) | AP^r_vol: 17.0 -> 14.6 | Hurt |
| use bounding boxes for thing classes rather than image-level tags for stuff classes | weakly supervised model · Cityscapes val | IoU: 60.2 -> 68.2 | Improved |
Closest papers by meaning
- Pointly-Supervised Panoptic Segmentation (ECCV 2022 · similarity 0.92)
- Simple Does It: Weakly Supervised Instance and Semantic Segmentation (CVPR 2017 · similarity 0.92)
- Toward Joint Thing-and-Stuff Mining for Weakly Supervised Panoptic Segmentation (CVPR 2021 · similarity 0.92)
- Fully Convolutional Networks for Panoptic Segmentation with Point-based Supervision (TPAMI 2022 · similarity 0.91)
- Universal Weakly Supervised Segmentation by Pixel-to-Segment Contrastive Learning (ICLR 2021 · similarity 0.91)
- WISH: Weakly Supervised Instance Segmentation using Heterogeneous Labels (CVPR 2025 · similarity 0.90)
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.
