MVREC: A General Few-shot Defect Classification Model Using Multi-View Region-Context

Authors: Shuai Lyu, Rongchen Zhang, Zeqi Ma, Fangjian Liao, Dongmei Mo, Waikeung Wong

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments conducted on MVTec-FS and four additional datasets demonstrate its effectiveness in general defect classification and its ability to incorporate contextual information to improve classification performance.
Researcher Affiliation Academia 1 The Hong Kong Polytechnic University, Hong Kong SAR, China 2 Laboratory for Artificial Intelligence in Design, Hong Kong SAR, China EMAIL EMAIL
Pseudocode No The paper describes methods like Zip-Adapter and Zip-Adapter-F with mathematical formulations (Equations 1-6) and a framework diagram (Figure 2), but no explicitly labeled 'Pseudocode' or 'Algorithm' block is provided.
Open Source Code Yes Code https://github.com/Shuai LYU/MVREC
Open Datasets Yes We also introduce MVTec-FS, a new FSDMC benchmark based on MVTec AD... We evaluated MVREC on several public datasets, including: 1) NEU-DET (He et al. 2020)... 2) PCB Defect Dataset (Huang and Wei 2019)... 3) Magnetic Tile Surface Defects (MTD) (Huang, Qiu, and Yuan 2020)... and 4) AITEX Fabric Defect (Silvestre-Blanes et al. 2019)...
Dataset Splits Yes For each defect type, 50% of the defects are used as the training set to sample the support set, while the remaining 50% constitute the testing set (query set).
Hardware Specification No The paper does not explicitly state any specific hardware details such as GPU models, CPU types, or memory specifications used for conducting the experiments.
Software Dependencies No The paper mentions the Adam W optimizer and Alpha CLIP model, but does not provide specific version numbers for software dependencies such as programming languages, libraries, or frameworks.
Experiment Setup Yes For Alpha CLIP, we selected the Vi TL/14 (Dosovitskiy et al. 2020) backbone. The MVREC visual feature used three scales, representing the commonly used large, medium, and small settings. The number of offsets was set to 9... We set β to 32 for the Zip-Adapter and 1 for the Zip-Adapter-F classifiers. When training the Zip-Adapter F, we used the Adam W optimizer with a learning rate of 0.0001. The model was updated for 500 iterations... For the triplet loss item, the hyperparameters α and λ were set to 0.5 and 4, respectively.