Research map / Learning theory and algorithms / Optimal transport
Bridging Vision and Language Concepts through Optimal Transport Semantic Flow
Chenyang Zhang, Anqi Dong, Guangming Zhu, Nuoye Xiong, Siyuan Wang, Lin Mei, Liang Zhang
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 |
|---|---|---|---|
| replace similarity-based concept inference with classical OT (ablation) | Vanilla-CBM · ImageNet | accuracy: 79.17 -> 80.42 | Improved |
| replace similarity-based concept inference with classical OT (ablation) | Vanilla-CBM · CIFAR-100 | accuracy: 80.04 -> 85.23 | Improved |
| use Vision-Language OT instead of classical OT (ablation) | Vanilla-CBM + OT · CUB | accuracy: 80.31 -> 84.31 | Improved |
| use Vision-Language OT instead of classical OT (ablation) | Vanilla-CBM + OT · Places365 | accuracy: 47.7 -> 50.59 | Improved |
| add inverse OT learned cost (ablation) | Vanilla-CBM + VLOT · CUB | accuracy: 84.31 -> 87.44 | Improved |
| add inverse OT learned cost (ablation) | Vanilla-CBM + VLOT · ImageNet | accuracy: 82.68 -> 84.87 | Improved |
| add semantic flow with velocity-based activation (ablation) | Vanilla-CBM + VLOT + IoT · CUB | accuracy: 87.44 -> 89.92 | Improved |
| add semantic flow with velocity-based activation (ablation) | Vanilla-CBM + VLOT + IoT · AwA2 | accuracy: 96.56 -> 98.88 | Improved |
| add semantic flow with velocity-based activation (ablation) | Vanilla-CBM + VLOT + IoT · Places365 | accuracy: 52.99 -> 55.13 | Improved |
| use learned IoT cost instead of cosine distance or W2^2 (ablation) | OT alignment · Cost learning (TRE) | TRE: 2.371 -> 0.824 | Improved |
| use UOT instead of balanced OT in learned flow matching (ablation) | IoT + OT Flow Matching · Semantic flow modeling | VMSE: 1.782 -> 1.000 | Improved |
| use UOT instead of balanced OT in learned flow matching (ablation) | IoT + OT Flow Matching · Semantic flow modeling | NPE: 0.197 -> 0.021 | Improved |
Closest papers by meaning
- Discovering Fine-Grained Visual-Concept Relations by Disentangled Optimal Transport Concept Bottleneck Models (CVPR 2025 · similarity 0.88)
- V2C-CBM: Building Concept Bottlenecks with Vision-to-Concept Tokenizer (AAAI 2025 · similarity 0.87)
- Hyperbolic Concept Bottleneck Models (NeurIPS 2026 · similarity 0.87)
- Hybrid Concept Bottleneck Models (CVPR 2025 · similarity 0.87)
- Coarse-to-Fine Concept Bottleneck Models (NeurIPS 2024 · similarity 0.86)
- Beyond Scalars: Concept-Based Alignment Analysis in Vision Transformers (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.
