This project is for the paper "Training Confidence-Calibrated Classifier for Detecting Out-of-Distribution Samples". Some codes are from odin-pytorch.
It is tested under Ubuntu Linux 16.04.1 and Python 2.7 environment, and requries Pytorch package to be installed:
- Pytorch: Only GPU version is available.
We use download links of two out-of-distributin datasets from odin-pytorch:
run_cross_entropy.sh: train the models using standard cross entropy loss.run_joint_confidence.sh: train the models using joint confidence loss.
test.sh--dataset --out_dataset --pre_trained_net
--dataset = name of in-distribution (svhn or cifar10)
--out_dataset = name of out-of-distribution (svhn, cifar10, lsun or imagenet)
--pre_trained_net = path to pre_trained_net