PyTorch implementation for "Classification-Reconstruction Learning for Open-Set Recognition" CVPR 2019. It is important to note that this repository borrows heavily from the repo - https://github.com/abhijitbendale/OSDN as far as fitting weibull distribution and computing open max scores goes. It also borrows from the original repo for the paper - https://nae-lab.org/~rei/research/crosr (which is implemented in chainer). I think thsi repository is much cleaner and flexible to accomodate newer datasets and model architectures. Some part of code works on python 3 and 2 whereas the other portion only works on python 2.7. Please follow the following steps to successfully run the code.

Compile LibMR and python interface to LibMR using following commands. For pythong interfaces to work, you would require Cython to be pre-installed on your machine

cd libMR/
chmod +x compile.sh
./compile.sh

python train_net.py

python get_model_features.py

python MAV_Compute.py

python compute_distances.py

python compute_openmax.py

The AUROC score for CIFAR-10 with 6/4 split is 71.23.