Adaptive Multi-prompt Contrastive Network for Few-shot Out-of-distribution Detection
Authors: Xiang Fang, Arvind Easwaran, Blaise Genest
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show that AMCN outperforms other state-of-the-art works. |
| Researcher Affiliation | Academia | 1Energy Research Institute @ NTU, Interdisciplinary Graduate Programme, Nanyang Technological University, Singapore 2College of Computing and Data Science, Nanyang Technological University, Singapore 3CNRS and CNRS@CREATE, IPAL IRL 2955, France and Singapore. Correspondence to: Xiang Fang <EMAIL>. |
| Pseudocode | No | The paper describes the methodology using textual explanations and mathematical equations, but does not include any clearly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | Yes | Codes are available in Github. |
| Open Datasets | Yes | For fair comparison, we follow MOS (Huang & Li, 2021) and MCM (Ming et al., 2022) to utilize Image Net1k (Deng et al., 2009) as ID set and a subset of i Naturalist (Horn et al., 2018), PLACES (Zhou et al., 2018), TEXTURE (Cimpoi et al., 2014) and SUN (Xiao et al., 2010) as OOD set. |
| Dataset Splits | Yes | For the K-shot OOD detection task, it aims to use only K labeled ID images from each class for model training, and to test on the complete test set for OOD detection. ... For the few-shot setting (Ye et al., 2020), following previous works (Miyai et al., 2023; Ye et al., 2020), we try different shots (1, 2, 4, 8, 16). |
| Hardware Specification | No | The computational work for this article was (fully/partially) performed on resources of the National Supercomputing Centre, Singapore (https://www.nscc.sg). This statement indicates the use of a supercomputing center but does not provide specific hardware details like GPU/CPU models or memory. |
| Software Dependencies | No | The paper mentions using 'CLIP-Vi T-B/16 (Radford et al., 2021) as the pretrained model' and 'Adam W (Loshchilov & Hutter, 2019) as the optimizer', but does not specify software versions for libraries, frameworks, or programming languages (e.g., Python, PyTorch versions). |
| Experiment Setup | Yes | For the parameters, we set α1 = 0.4, α2 = 0.2, α3 = 0.8, θ = 0.8, τ = 1.0, γ = 0.7, P = 1, S = 50, Z = 50. We set Adam W (Loshchilov & Hutter, 2019) as the optimizer, the learning rate of 0.003, the batch size as 64, the token length as 16 and the training epoch as 100. |