Research map / Continual learning / Domain and prompt-based
Hyperbolic Prototype Learning with Uncertainty-Aware Consistency for Continual Test-Time Segmentation
Siddhant Gole, Akash Pal, Amit More, S Divakar Bhat, Subhasis Chaudhuri, Biplab Banerjee
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 |
|---|---|---|---|
| use hyperbolic (Poincare ball) space instead of Euclidean for prototype segmentation (ablation) | HyperProtoSeg · Cityscapes (source val) | mIoU: 78.07 -> 79.96 | Improved |
| use hyperbolic geometry to reduce error accumulation over 10 CTTA rounds (ablation) | HyperProtoSeg · Cityscapes->ACDC | mIoU degradation round 1 to 10: -1.46 -> -0.62 | Improved |
| add prototypical framework to Euclidean SegFormer (ablation) | Euclidean SegFormer · Cityscapes (source val) | mIoU: 76.74 -> 78.07 | Improved |
| add regularizer Lreg on Euclidean feature norms (ablation) | HyperProtoSeg · IDD (source val) | mIoU: 70.17 -> 72.29 | Improved |
| add prototype separation loss Lsep (ablation) | HyperProtoSeg · IDD (source val) | mIoU: 72.29 -> 74.34 | Improved |
| use consistency-based uncertainty instead of MC dropout (ablation) | HBCA · IDD->IDD-AW CTTA | mIoU: 52.78 -> 57.12 | Improved |
| use consistency-based uncertainty instead of hyperbolic norm (ablation) | HBCA · IDD->IDD-AW CTTA | mIoU: 53.27 -> 57.12 | Improved |
| add directional boundary loss on uncertain pixels (ablation) | HBCA · IDD->IDD-AW CTTA | mIoU: 56.24 -> 57.12 | Improved |
| apply geodesic distance minimization (Lcore) on uncertain pixels (ablation) | HBCA · IDD->IDD-AW CTTA | mIoU: 56.24 -> 56.01 | Hurt |
| apply geodesic distance minimization (Lcore) on uncertain pixels (ablation) | HBCA · Cityscapes->ACDC CTTA | mIoU: 56.34 -> 55.82 | Hurt |
| HBCA vs best prior CTTA (Hybrid TTA) on IDD->IDD-AW | HBCA · IDD->IDD-AW CTTA | mean mIoU: 54.91 -> 57.12 | Improved |
| entropy minimization (TENT) as CTTA method | HyperProtoSeg source model · IDD->IDD-AW CTTA | mean mIoU: 51.31 -> 51.45 | No clear effect |
Closest papers by meaning
- Hyperbolic Active Learning for Semantic Segmentation under Domain Shift (ICML 2024 · similarity 0.88)
- Hyperbolic Uncertainty-Aware Few-Shot Incremental Point Cloud Segmentation (CVPR 2025 · similarity 0.86)
- Learning Along the Arrow of Time: Hyperbolic Geometry for Backward-Compatible Representation Learning (ICML 2025 · similarity 0.86)
- Robust Hyperbolic Learning with Curvature-Aware Optimization (NeurIPS 2025 · similarity 0.86)
- Test-Time Adaptation of Vision-Language Models for Open-Vocabulary Semantic Segmentation (NeurIPS 2025 · similarity 0.86)
- Unsupervised Feature Learning with Emergent Data-Driven Prototypicality (CVPR 2024 · 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.
