Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

Long-Tailed Out-of-Distribution Detection: Prioritizing Attention to Tail

Authors: Yina He, Lei Peng, Yongcun Zhang, Juanjuan Weng, Shaozi Li, Zhiming Luo

AAAI 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments verified that our method outperforms the current stateof-the-art methods on various benchmarks. Experiments Experiment Settings Datasets: We conduct experiments on widely used datasets, i.e., CIFAR10-LT, CIFAR100-LT (Cao et al. 2019), and Image Net-LT (Liu et al. 2019) as ID training sets (Din).
Researcher Affiliation Academia 1Department of Artificial Intelligence, Xiamen University, Xiamen, China 2College of Information Science and Technology, Jinan University, Guangzhou, China 3Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University
Pseudocode No More detailed configuration information is presented in Appendix C, and the full algorithm of our PATT is described in Appendix A.
Open Source Code Yes Code https://github.com/Ina R-design/PATT.
Open Datasets Yes Datasets: We conduct experiments on widely used datasets, i.e., CIFAR10-LT, CIFAR100-LT (Cao et al. 2019), and Image Net-LT (Liu et al. 2019) as ID training sets (Din). The standard CIFAR10, CIFAR100, and Image Net test sets are used as ID test sets (Dtest in ). Following PASCL (Wang et al. 2022), we utilize 300,000 samples from Tiny Images80M (Torralba, Fergus, and Freeman 2008) as the surrogate OOD training data for CIFAR10/100-LT and Image Net Extra as the surrogate OOD training for Image Net-LT. We set the default imbalance ratio to 100 for CIFAR10/100-LT. For OOD test data, we use Textures (Cimpoi et al. 2014), SVHN (Netzer et al. 2011), Tiny Image Net (Le and Yang 2015), LSUN (Yu et al. 2015), and Places365 (Zhou et al. 2017) introduced in the SC-OOD benchmark (Yang et al. 2021) as Dout test for CIFAR10/100-LT.
Dataset Splits Yes Datasets: We conduct experiments on widely used datasets, i.e., CIFAR10-LT, CIFAR100-LT (Cao et al. 2019), and Image Net-LT (Liu et al. 2019) as ID training sets (Din). The standard CIFAR10, CIFAR100, and Image Net test sets are used as ID test sets (Dtest in ). ... For OOD test data, we use Textures (Cimpoi et al. 2014), SVHN (Netzer et al. 2011), Tiny Image Net (Le and Yang 2015), LSUN (Yu et al. 2015), and Places365 (Zhou et al. 2017) introduced in the SC-OOD benchmark (Yang et al. 2021) as Dout test for CIFAR10/100-LT.
Hardware Specification No No specific hardware details (GPU/CPU models, memory) are mentioned in the provided text.
Software Dependencies No Following PASCL (Wang et al. 2022), we use Res Net-18 (He et al. 2016) as our backbone and perform the experiments using the Adam optimizer (Kingma and Ba 2014) with an initial learning rate 1 10 3 for experiments on CIFAR10/100-LT. For Image Net-LT, we use Res Net-50 (He et al. 2016) as our backbone and train the model using the SGD optimizer with an initial learning rate of 0.1.
Experiment Setup Yes Configuration: Following PASCL (Wang et al. 2022), we use Res Net-18 (He et al. 2016) as our backbone and perform the experiments using the Adam optimizer (Kingma and Ba 2014) with an initial learning rate 1 10 3 for experiments on CIFAR10/100-LT. For Image Net-LT, we use Res Net-50 (He et al. 2016) as our backbone and train the model using the SGD optimizer with an initial learning rate of 0.1. All experiments are conducted by training the model for 100 epochs, with a batch size of 128.