LoCoOp: Few-Shot Out-of-Distribution Detection via Prompt Learning

Authors: Atsuyuki Miyai, Qing Yu, Go Irie, Kiyoharu Aizawa

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on the large-scale Image Net OOD detection benchmarks demonstrate the superiority of our Lo Co Op over zero-shot, fully supervised detection methods and prompt learning methods.
Researcher Affiliation Collaboration Atsuyuki Miyai1 Qing Yu1,2 Go Irie3 Kiyoharu Aizawa1 1The University of Tokyo 2LY Corporation 3Tokyo University of Science
Pseudocode No The paper includes diagrams and descriptions of the method but no formal pseudocode or algorithm blocks.
Open Source Code Yes The code is available via https: //github.com/Atsu Miyai/Lo Co Op.
Open Datasets Yes Datasets. We use the Image Net-1K dataset [5] as the ID data. For OOD datasets, we adopt the same ones as in [18], including subsets of i Naturalist [47], SUN [51], Places [58], and TEXTURE [4].
Dataset Splits Yes For the few-shot training, we follow the few-shot evaluation protocol adopted in CLIP [37] and Co Op [60], using 1, 2, 4, 8, and 16 shots for training, respectively, and deploying models in the full test sets.
Hardware Specification Yes We use a single Nvidia A100 GPU for all experiments.
Software Dependencies No The paper mentions using 'CLIP-Vi T-B/16 models' and refers to CLIP, but it does not list specific software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions).
Experiment Setup Yes Other hyperparameters (e.g., training epoch=50, learning rate=0.002, batch size=32, and token lengths N=16) are the same as those of Co Op [60].