Research map / Label-efficient and robust learning / Recognition and long-tail
Robust Multi-View Fusion via Prototype-Anchored Unbalanced Optimal Transport
Zhang Han, Xingwen Zhao, HUI LI
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 UOT fusion instead of cross-attention fusion (ablation) | UOT-Fusion · NTU 60 X-Sub, 40% missing | Top-1 accuracy (%): 83.0 -> 90.6 | Improved |
| use unbalanced OT (tau=0.5) instead of near-balanced OT (tau=10) (ablation) | UOT-Fusion · NTU 60 X-Sub, 40% missing | Top-1 accuracy (%): 87.0 -> 90.6 | Improved |
| add shared prototype support (ablation) | UOT-Fusion · NTU 60 X-Sub, 40% missing | Top-1 accuracy (%): 87.4 -> 90.6 | Improved |
| add mass-dependent modality weights lambda (ablation) | UOT-Fusion · NTU 60 X-Sub, 40% missing | Top-1 accuracy (%): 89.3 -> 90.6 | Improved |
| increase atoms and prototypes from J=K=16 to 32 (ablation) | UOT-Fusion · NTU 60 X-Sub, 40% missing | Top-1 accuracy (%): 90.6 -> 90.9 | No clear effect |
| set tau very small (0.05) (ablation) | UOT-Fusion · NTU 60 X-Sub, 40% missing | Top-1 accuracy (%): 90.6 -> 85.7 | Hurt |
| set tau large (2.0) (ablation) | UOT-Fusion · NTU 60 X-Sub, 40% missing | Top-1 accuracy (%): 90.6 -> 88.1 | Hurt |
| use UOT fusion instead of RCMCL in the clean setting | UOT-Fusion · NTU 60 X-Sub, 0% missing | Top-1 accuracy (%): 96.2 -> 96.0 | No clear effect |
| use UOT fusion instead of RCMCL under 80% missing | UOT-Fusion · NTU 60 X-Sub, 80% missing | Top-1 accuracy (%): 70.8 -> 78.3 | Improved |
| replace cosine distance cost with normalized squared Euclidean (ablation) | UOT-Fusion · NTU 60 | Top-1 accuracy (%): ±0.3% accuracy | No clear effect |
| use UOT fusion under RGB latent noise sigma=2.0 | UOT-Fusion · NTU 60 X-Sub, RGB noise | Top-1 accuracy (%): 83.1 -> 89.7 | Improved |
Closest papers by meaning
- Differentiable Hierarchical Optimal Transport for Robust Multi-View Learning (TPAMI 2022 · similarity 0.86)
- Deep Incomplete Multi-view Learning via Cyclic Permutation of VAEs (ICLR 2025 · similarity 0.84)
- Rethinking Multi-view Representation Learning via Distilled Disentangling (CVPR 2024 · similarity 0.84)
- Trusted Unified Feature-Neighborhood Dynamics for Multi-View Classification (AAAI 2025 · similarity 0.83)
- SeqMvRL: A Sequential Fusion Framework for Multi-view Representation Learning (CVPR 2025 · similarity 0.83)
- DVANet: Disentangling View and Action Features for Multi-View Action Recognition (AAAI 2024 · similarity 0.83)
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.
