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].
Few Shot Part Segmentation Reveals Compositional Logic for Industrial Anomaly Detection
Authors: Soopil Kim, Sion An, Philip Chikontwe, Myeongkyun Kang, Ehsan Adeli, Kilian M. Pohl, Sang Hyun Park
AAAI 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on the public benchmark MVTec LOCO AD reveal our method achieves 98.1% AUROC in LA detection vs. 89.6% from competing methods. |
| Researcher Affiliation | Academia | 1Robotics and Mechatronics Engineering, DGIST, Daegu, Korea 2AI Graduate School, DGIST, Daegu, Korea 3Stanford University, Stanford, CA 94305, USA |
| Pseudocode | No | The paper does not contain a pseudocode block or a clearly labeled algorithm. |
| Open Source Code | No | The paper does not include any explicit statement about making the source code available or provide a link to a code repository. |
| Open Datasets | Yes | We evaluated our method on MVTec LOCO AD dataset (Bergmann et al. 2022) |
| Dataset Splits | No | For each category, 351/335/372/360/360 normal images were used for training and 275/330/310/341/312 images for testing following the setting of the comparison methods. The paper mentions training and testing splits but does not specify a separate validation split for their own experiments, stating they use training data for adaptive scaling instead of a validation dataset. |
| Hardware Specification | Yes | For training, we used an Adam W optimizer with a learning rate 0.001 and batch size of 5 per iteration on an NVIDIA RTX A5000 GPU workstation. |
| Software Dependencies | No | The paper mentions software components like 'Adam W optimizer' and 'Wide Res Net101' but does not provide specific version numbers for these or any other software dependencies needed for replication. |
| Experiment Setup | Yes | For training, we used an Adam W optimizer with a learning rate 0.001 and batch size of 5 per iteration on an NVIDIA RTX A5000 GPU workstation. The model was first trained for 50 epochs using only LCE and LDice. After warming up with the supervised loss, the model is trained using Eq.(1) for additional 50 epochs. As LDice is usually larger than the other losses, hyper-parameters λ1, λ2, and λ3 were set as 10. |