rmagesh148/combood

# Method

• The 4 most important files for this paper are test_ood.py, utils.py, data.py, confidenciator.py

• Download the OpenOOD (Git repo: https://github.com/Jingkang50/OpenOOD/tree/main) datasets and checkpoints from this link: https://entuedu-my.sharepoint.com/:f:/g/personal/jingkang001_e_ntu_edu_sg/Eso7IDKUKQ9AoY7hm9IU2gIBMWNnWGCYPwClpH0TASRLmg?e=kMrkVQ

• Run the test_ood.py file to check results.

• Make sure you provide necessary directory for each dataset and checkpoint(pretrained models) before running the code.

• Change the directory of "OpenOOD " (/confidence-magesh_MR/confidence-magesh/OpenOOD) folder inside load.py (/confidence-magesh/models/load.py) script.

• Similarly change the directory of "OpenOOD" folder inside data.py (/confidence-magesh/data.py) script.

# Cifar10:

• Provide necessary directories of pretrained OpenOOD checkpoint models inside the script: /confidence-magesh/OpenOOD/openood_id_ood_and_model_cifar10.py

• Provide necessary directories of OpenOOD datasets inside the files: /home/saiful/confidence_icdb/confidence-magesh/OpenOOD/configs/datasets/cifar10/cifar10.yml and /home/saiful/confidence_icdb/confidence-magesh/OpenOOD/configs/datasets/cifar10/cifar10_ood.yml.

Please follow the similar approach to run it with mnist, cifar100, and imagenet. You need to provide directory of the OpenOOD datasets and checkpoints inside: /confidence-magesh/OpenOOD/openood_id_ood_and_model_mnist.py,
/confidence-magesh/OpenOOD/openood_id_ood_and_model_cifar100.py,
and confidence-magesh/OpenOOD/openood_id_ood_and_model_imagenet.py files.

• The results can be found inside the following directories: for mnist : /confidence-magesh/results/mnist_lenet/knn/ for cifar10: /confidence-magesh/results/cifar10_resnet/knn/ for cifar100: /confidence-magesh/results/cifar100_resnet/knn/ for imagenet: /confidence-magesh/results/imagenet_resnet50/knn/ for document: /confidence-magesh/results/document_resnet50_docu/knn/

# Document dataset:

• Download the dataset from this link https://adamharley.com/rvl-cdip/
• Preprocess the dataset folder directories following this link https://github.com/MdSaifulIslamSajol/mobilenet_image_classification_with_document_dataset/blob/main/make_classwise_subfolders_rvl_cdip.py
• The processed dataset can also directly be downloaded from this link https://lsu.box.com/s/x71r0eiagqgbqxei50ghbk9cldslxr34
• The pretrained checkpoints of Resnet50 for document dataset can be found on this directory: /confidence-magesh/document classification/saved trained models/resnet50_checkpoints/resnet50_acc0.9_epoch40_on_319837_trainimages_load.ckpt"
• The OOD datasets for document dataset can be found on this link: https://github.com/gxlarson/rvl-cdip-ood
• Now provide directory of the OpenOOD datasets and checkpoints inside: confidence-magesh/document_id_ood_n_model_loader.py script .

Citation

If you find our repository useful for your research, please consider citing our paper:

# v1.0
@Book{magesh2024combood,
	author = {Magesh Rajasekaran and Md Saiful Islam Sajol and Frej Berglind and Supratik Mukhopadhyay and Kamalika Das},
	title = {COMBOOD: A Semiparametric Approach for Detecting Out-of-distribution Data for Image Classification},
	booktitle = {Proceedings of the 2024 SIAM International Conference on Data Mining (SDM)},
	pages = {643-651},
	year = {2024},
	doi = {10.1137/1.9781611978032.74},
	URL = {https://epubs.siam.org/doi/abs/10.1137/1.9781611978032.74}
}