Official PyTorch implementation of Class Specific Semantic Reconstruction for Open Set Recognition.

Before training, please setup dataset directories in dataset.py:

DATA_PATH = ''          # path for cifar10, svhn
TINYIMAGENET_PATH = ''  # path for tinyimagenet
LARGE_OOD_PATH = ''     # path for ood datasets, e.g., iNaturalist in imagenet experiment
IMAGENET_PATH = ''      # path for imagenet-1k datasets

To train models from scratch, run command:

python main.py --gpu 0 --ds {DATASET} --config {MODEL} --save {SAVING_NAME} --method cssr

Command options:

• DATASET: Experiment configuration file, specifying datasets and random splits, e.g., ./exps/$dataset/spl_$s.json.
• MODEL: OSR model configuration file, specifying model parameters, e.g., ./configs/$model/$dataset.json. $model includes linear/pcssr/rcssr, which corresponds to the baseline and the proposed model.

Or simply run bash file sh run.sh to run all experiments simultaneously.

To train models by finetuning pretrained backbones, like experiments for imagenet-1k, run command:

python main.py --gpu 0 --ds ./exps/imagenet/vs_inaturalist.json --config ./configs/rcssr/imagenet.json --save imagenet1k_rcssr --method cssr_ft

Add --test on training commands to restore and evaluate a pretrained model on specified data setup, e.g.,

python main.py --gpu 0 --ds {DATASET} --config {MODEL} --save {SAVING_NAME} --method cssr --test

With models trained by sh run.sh, script collect_metrics.py helps collect and present experimental results: python collect_metrics.py

@ARTICLE{9864101,
  author={Huang, Hongzhi and Wang, Yu and Hu, Qinghua and Cheng, Ming-Ming},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  title={Class-Specific Semantic Reconstruction for Open Set Recognition},
  year={2022},
  doi={10.1109/TPAMI.2022.3200384}
}