The code will be sorted and published soon.
FCTL is developed based on torch==1.13.0 torchvision==0.14.0 pytorch-lightning==2.2.1
We provide the detail about the benchmark datasets in data/.
| Dataset | Backbone | AUROC | Log | Weight (Backup) |
|---|---|---|---|---|
| MNIST | ResNet101 | 0.999 | 🤗split1 split2 split3 split4 split5 | |
| SVHN | ResNet101 | 0.972 | ||
| CIFAR10 | ResNet101 | 0.948 | ||
| CIFAR+10 | ResNet101 | 0.990 | ||
| CIFAR+50 | ResNet101 | 0.979 | ||
| TinyImageNet | ResNet101 | 0.951 | ||
| CUB | ResNet101 | 0.812 | split1_log | |
| Aircraft | ResNet101 | 0.853 | split1_log | 🤗split1 |
| MNIST | ViT/B16 | 0.983 | split1_log | |
| SVHN | ViT/B16 | 0.974 | ||
| CIFAR10 | ViT/B16 | 0.987 | 🤗split1 split2 split3 split4 split5 | |
| CIFAR+10 | ViT/B16 | 0.995 | 🤗split1 split2 split3 split4 split5 | |
| CIFAR+50 | ViT/B16 | 0.991 | 🤗split1 split2 split3 split4 split5 | |
| TinyImageNet | ViT/B16 | 0.978 | 🤗split1 split2 split3 split4 split5 | |
| CUB | ViT/B16 | 0.895 | 🤗split1 | |
| Aircraft | ViT/B16 | 0.817 | split1_log | |
| CIFAR10-ood | ResNet101 | 🤗split1 |
https://drive.google.com/file/d/12apkcoi-YQRP5IK9MVBQ_r-zFw6byLyE/view?usp=drive_link
chmod +x _scripts/*/test*.shThe code will be sorted and published soon.