Image Classification on CIFAR100 (Lower and Higher Level Accuracy)
81.54Lower Level AccuracyHIL-CBM(Features)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HIL-CBM(Features)Backbone=ViT, Sparse=Yes2026.04 | 81.54 | 86.12 | |
| HIL-CBM(Class names)Backbone=ViT, Sparse=Yes2026.04 | 79.24 | 86.04 | |
| StandardBackbone=ResNet, Sparse=No2026.04 | 70.1 | — | |
| HIL-CBMBackbone=ResNet, Sparse=No2026.04 | 69.91 | 72.36 | |
| HIL-CBMBackbone=ResNet, Sparse=Yes2026.04 | 69.5 | 73.55 | |
| HybridCBMBackbone=ResNet, Sparse=Yes2026.04 | 68.38 | — | |
| LF-CBMBackbone=ResNet, Sparse=Yes2026.04 | 65.13 | — | |
| Standard HierarchicalBackbone=ResNet, Sparse=Yes2026.04 | 58.6 | 72.39 | |
| StandardBackbone=ResNet, Sparse=Yes2026.04 | 58.34 | — | |
| P-CBMBackbone=ResNet, Sparse=Yes2026.04 | 43.2 | — |