Research map / Label-efficient and robust learning / Domain adaptation and generalization
MC-PanDA++: Simpler, Stronger, and More Robust Domain-Adaptive Panoptic Segmentation
Ivan Martinović, Josip Šarić, Yuki M. Asano, Siniša Šegvić
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 MLS (mask-wide loss scaling) to baseline consistency (ablation) | MC-PanDA (Swin-B) · Synthia->Cityscapes | PQ16: 39.6 -> 44.6 | Improved |
| add CBPF (confidence-based point filtering) to baseline consistency (ablation) | MC-PanDA (Swin-B) · Synthia->Cityscapes | PQ16: 39.6 -> 44.1 | Improved |
| add CBPF on top of MLS (ablation) | MC-PanDA (Swin-B) · Synthia->Vistas | PQ16: 37.1 -> 38.7 | Improved |
| use mask-wide loss scaling instead of image-wide loss scaling (ablation) | MC-PanDA (Swin-B) · Synthia->Cityscapes | PQ16: 42.1 -> 47.4 | Improved |
| use all-mask confidence instead of per-mask confidence in CBPF (ablation) | MC-PanDA (Swin-B) · Synthia->Cityscapes | PQ16: 44.7 -> 47.4 | Improved |
| use CBPF instead of random point sampling (ablation) | MC-PanDA (Swin-B) · Synthia->Cityscapes | PQ16: 44.2 -> 47.4 | Improved |
| use self-supervised DINO initialization instead of supervised ImageNet-1k (ablation) | MC-PanDA ViT-B/16 · Synthia->Cityscapes | PQ: +4.2 PQ | Improved |
| replace Swin-B with self-supervised DINOv2-B initialization (ablation) | MC-PanDA · Synthia->Vistas | PQ16: +8.5 PQ16 | Improved |
| add baseline consistency alone with DINOv2-B (ablation) | MC-PanDA (DINOv2-B) · Synthia->Vistas | PQ16: 33.7 -> 33.6 | No clear effect |
| add baseline consistency alone with DINOv2-B (ablation) | MC-PanDA (DINOv2-B) · Cityscapes->ACDC | PQ19: 48.7 -> 43.3 | Hurt |
| add MLS and CBPF to baseline consistency with DINOv2-B (ablation) | MC-PanDA (DINOv2-B) · Cityscapes->ACDC | PQ19: 43.3 -> 55.5 | Improved |
| use UrbanSyn instead of Synthia as source domain | MC-PanDA (DINOv2-B) · Cityscapes (16 shared classes) | PQ16: 48.5 -> 59.8 | 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.
- MC-PanDA — MC-PanDA: Mask Confidence for Panoptic Domain Adaptation (ECCV 2024) · better on 6 of 6 shared results
- EDAPS — EDAPS: Enhanced Domain-Adaptive Panoptic Segmentation (ICCV 2023) · better on 4 of 4 shared results
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
- PanSt3R: Multi-view Consistent Panoptic Segmentation (ICCV 2025 · similarity 0.90)
- MC-PanDA: Mask Confidence for Panoptic Domain Adaptation (ECCV 2024 · similarity 0.89)
- Cross-View Regularization for Domain Adaptive Panoptic Segmentation (CVPR 2021 · similarity 0.88)
- EDAPS: Enhanced Domain-Adaptive Panoptic Segmentation (ICCV 2023 · similarity 0.87)
- A Generalist Framework for Panoptic Segmentation of Images and Videos (ICCV 2023 · similarity 0.86)
- Combinatorial Optimization for Panoptic Segmentation: A Fully Differentiable Approach (NeurIPS 2021 · similarity 0.86)
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.
