Research map / Label-efficient and robust learning / Recognition and long-tail
Exchangeable Deep Neural Networks for Set-to-Set Matching and Learning
Yuki Saito, Takuma Nakamura, Hirotaka Hachiya, Kenji Fukumizu
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 K-pair-set loss instead of triplet loss (ablation) | Cross Affinity · Market-1501 Group average | accuracy (%): 45.5 -> 87.0 | Improved |
| add cross-set feature transformation (CSeFT) (ablation) | Cross Affinity · Market-1501 Group average | accuracy (%): 82.8 -> 87.0 | Improved |
| use two CSeFT layers instead of one (ablation) | Cross Affinity · Market-1501 Group average | accuracy (%): 86.0 -> 87.0 | Improved |
| use eight heads instead of one (ablation) | Cross Affinity · Market-1501 Group average | accuracy (%): 85.9 -> 87.0 | Improved |
| add encoder (self-attention block) (ablation) | Cross Affinity · Market-1501 Group average | accuracy (%): 85.7 -> 87.0 | Improved |
| use multiple cross-similarity functions instead of single CS (ablation) | Cross Affinity · Market-1501 Group average | accuracy (%): 86.0 -> 87.0 | Improved |
| add ReLU in cross-similarity (ablation) | Cross Affinity · Market-1501 Group average | accuracy (%): 85.0 -> 87.0 | Improved |
| use mCS instead of max pooling (ablation) | Cross Affinity · Market-1501 Group average | accuracy (%): 86.1 -> 87.0 | Improved |
| use mCS instead of cosine similarity metric (ablation) | Cross Affinity · Market-1501 Group average | accuracy (%): 53.1 -> 87.0 | Improved |
| add set-aug data augmentation (ablation) | Cross Affinity · Road Group CMC-1 | CMC-1 (%): 45.2 -> 84.0 | Improved |
| add image augmentation (img-aug) (ablation) | Cross Affinity · Road Group CMC-1 | CMC-1 (%): 45.2 -> 47.7 | No clear effect |
| add img-aug on top of set-aug (ablation) | Cross Affinity · Road Group CMC-1 | CMC-1 (%): 84.0 -> 81.7 | Hurt |
Closest papers by meaning
- Learn to Predict Sets Using Feed-Forward Neural Networks (TPAMI 2021 · similarity 0.84)
- Learning Deep Match Kernels for Image-Set Classification (CVPR 2017 · similarity 0.83)
- Set Prediction in the Latent Space (NeurIPS 2021 · similarity 0.83)
- DeepSetNet: Predicting Sets With Deep Neural Networks (ICCV 2017 · similarity 0.83)
- Joint Learning of Set Cardinality and State Distribution (AAAI 2018 · similarity 0.83)
- Unlocking Better Closed-Set Alignment Based on Neural Collapse for Open-Set Recognition (AAAI 2025 · similarity 0.80)
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.
