# 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/jingkang001entuedusg/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/openoodidoodandmodel_cifar10.py

• Provide necessary directories of OpenOOD datasets inside the files: /home/saiful/confidenceicdb/confidence-magesh/OpenOOD/configs/datasets/cifar10/cifar10.yml and /home/saiful/confidenceicdb/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/openoodidoodandmodelmnist.py,
/confidence-magesh/OpenOOD/openoodidoodandmodelcifar100.py,
and confidence-magesh/OpenOOD/openoodidoodandmodel_imagenet.py files.

• The results can be found inside the following directories: for mnist : /confidence-magesh/results/mnistlenet/knn/ for cifar10: /confidence-magesh/results/cifar10resnet/knn/ for cifar100: /confidence-magesh/results/cifar100resnet/knn/ for imagenet: /confidence-magesh/results/imagenetresnet50/knn/ for document: /confidence-magesh/results/documentresnet50docu/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/mobilenetimageclassificationwithdocumentdataset/blob/main/makeclasswisesubfoldersrvl_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/resnet50checkpoints/resnet50acc0.9epoch40on319837trainimages_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/documentidoodnmodel_loader.py script .

Citation

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

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} } ```