AdaNeg: Adaptive Negative Proxy Guided OOD Detection with Vision-Language Models

Authors: Yabin Zhang, Lei Zhang

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments across various benchmarks demonstrate the effectiveness of our approach, abbreviated as Ada Neg.
Researcher Affiliation Academia Yabin Zhang The Hong Kong Polytechnic University csybzhang@comp.polyu.edu.hk Lei Zhang The Hong Kong Polytechnic University cslzhang@comp.polyu.edu.hk
Pseudocode Yes Algorithm 1 Adaptive Negative Proxy Guided OOD Detection
Open Source Code Yes Codes are available at https://github.com/YBZh/Open OOD-VLM.
Open Datasets Yes We conduct extensive experiments with the large-scale Image Net-1k [9] as ID data. Following prior practice [26, 39, 27], four OOD datasets of i Naturalist [60], SUN [68], Places [78], and Textures [7] are evaluated.
Dataset Splits No The paper describes a training-free method and discusses hyperparameter tuning in Section 4.3, but it does not specify explicit training/validation/test dataset splits in the traditional sense for model validation during a training phase.
Hardware Specification Yes All experiments are conducted with a single Tesla V100 GPU.
Software Dependencies No The paper mentions using the visual encoder of VITB/16 pretrained by CLIP [48] but does not provide specific version numbers for software libraries or frameworks like PyTorch, TensorFlow, or Python versions.
Experiment Setup Yes For hyper-parameters, we adopt the memory length L=10, threshold γ=0.5 with the gap g=0.5 in Eq. 8, β=5.5 in Eq. 11, and λ=0.1 in Eq. 13 in all experiments.