Research map / Label-efficient and robust learning / Recognition and long-tail
T-vMF Similarity for Regularizing Intra-Class Feature Distribution
Takumi Kobayashi
Check 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 cosine similarity with t-vMF similarity (κ=64) | ResNet-10 · ImageNet-LT | top-1 error (%): 59.17 -> 56.31 | Improved |
| replace cosine similarity with t-vMF similarity (κ=16) | ResNet-50 · iNaturalist2018 | top-1 error (%): 32.20 -> 28.92 | Improved |
| replace cosine similarity with t-vMF similarity (κ=64) | ResNet-50 · iNaturalist2019 | top-1 error (%): 26.34 -> 25.08 | Improved |
| replace cosine similarity with t-vMF similarity (κ=16) | ResNet-10 · ImageNet-S | top-1 error (%): 52.69 -> 52.06 | Improved |
| replace cosine similarity with t-vMF similarity (κ=16) | ResNet-10 · ImageNet-SS | top-1 error (%): 66.41 -> 64.77 | Improved |
| replace cosine similarity with t-vMF similarity (κ=64) | ResNet-10 · ImageNet-N | top-1 error (%): 72.40 -> 69.19 | Improved |
| use t-vMF with κ=-0.3 rather than ArcFace | ResNet-50 · ImageNet | top-1 error (%): 23.28 -> 22.62 | Improved |
| increase κ from 4 to 64 (ablation) | t-vMF on ResNet-10 · ImageNet-LT | top-1 error (%): 59.17 -> 56.31 | Improved |
| replace t-vMF with vMF similarity at κ=16 (ablation) | ResNet-10 · ImageNet-LT | top-1 error (%): 57.30 -> 75.84 | Hurt |
| replace t-vMF with q-vMF similarity (ablation) | ResNet-10 · ImageNet-LT | top-1 error (%): 57.30 -> 59.05 | Hurt |
| set κ=0 to recover cosine similarity (ablation) | t-vMF on ResNet-10 · ImageNet-LT | top-1 error (%): 56.31 -> 61.32 | Hurt |
| learn a separate κ per class rather than fixing κ (ablation) | t-vMF on ResNet-10 · ImageNet-LT | top-1 error (%): 56.31 -> 60.39 | 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.
- ArcFace — ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019) · 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
- Utilizing Class Information for Deep Network Representation Shaping (AAAI 2019 · similarity 0.86)
- Class-Independent Regularization for Learning with Noisy Labels (AAAI 2023 · similarity 0.86)
- Tracing Representation Progression: Analyzing and Enhancing Layer-Wise Similarity (ICLR 2025 · similarity 0.85)
- ConR: Contrastive Regularizer for Deep Imbalanced Regression (ICLR 2024 · similarity 0.85)
- Stabilizing Sample Similarity in Representation via Mitigating Random Consistency (ICML 2025 · similarity 0.84)
- Learning from Neighbors: Category Extrapolation for Long-Tail Learning (CVPR 2025 · similarity 0.84)
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.
