Research map / Label-efficient and robust learning / Test-time adaptation
Mixture of Prototypes for Test-time Adaptive Segmentation
Guangrui Li, Zhengyu Zhu, Yongxin Ge
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 expert filtering (top-K) to L_min-trained gating (ablation) | MoE prototypes · Cityscapes-to-ACDC | Avg mIoU: 57.9 -> 61.7 | Improved |
| optimize gating with entropy minimization L_min alone (ablation) | MoE prototypes · Cityscapes-to-ACDC | Avg mIoU: 61.1 -> 57.9 | Hurt |
| add entropy maximization L_max on gating values (ablation) | MoE prototypes · Cityscapes-to-ACDC | Avg mIoU: 61.7 -> 62.0 | Improved |
| add confidence-guided instance weighting (ablation) | MoE prototypes · Cityscapes-to-ACDC | Avg mIoU: 62.0 -> 62.4 | Improved |
| K-means expert generation instead of farthest point sampling or autoencoder (ablation) | MoE prototypes · Cityscapes-to-ACDC | mIoU | Improved |
| use all 14 experts (K=14) instead of top-K filtering (ablation) | MoE prototypes · Cityscapes-to-Fog | mIoU | Hurt |
| set number of experts per class N to 14 (ablation) | MoE prototypes · Cityscapes-to-Fog | mIoU | Improved |
| use excessively large L_max coefficient beta (ablation) | MoE prototypes · Cityscapes-to-Fog | mIoU | Hurt |
| add MoE prototypes on SVDP baseline in continual TTA | SVDP · Cityscapes-to-ACDC continual TTA | mean mIoU: 61.3 -> 62.5 | Improved |
| domain-shared MoE with min-max optimization vs BECoTTA domain-specific experts | BECoTTA · Cityscapes-to-ACDC continual TTA | mean mIoU: 59.1 -> 62.5 | Improved |
| update model parameters during continual adaptation (TENT, DePT) | TENT and DePT · Cityscapes-to-ACDC continual TTA | mean mIoU: TENT 56.7 -> 55.0 over rounds | Hurt |
| method on ACDC rain vs SVDP | MoE prototypes · Cityscapes-to-ACDC TTA Rain | mIoU: 64.2 -> 63.5 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TENT — Tent: Fully Test-Time Adaptation by Entropy Minimization (ICLR 2021) · 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
- Improved Self-Training for Test-Time Adaptation (CVPR 2024 · similarity 0.88)
- Test-Time Adaptation of Vision-Language Models for Open-Vocabulary Semantic Segmentation (NeurIPS 2025 · similarity 0.88)
- Gradient Alignment Improves Test-Time Adaptation for Medical Image Segmentation (AAAI 2025 · similarity 0.87)
- Domain-Specific Block Selection and Paired-View Pseudo-Labeling for Online Test-Time Adaptation (CVPR 2024 · similarity 0.86)
- A Layer Selection Approach to Test Time Adaptation (AAAI 2025 · similarity 0.86)
- Backpropagation-Free Test-Time Adaptation via Probabilistic Gaussian Alignment (NeurIPS 2025 · 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.
