KairosAD: A SAM-Based Model for Industrial Anomaly Detection on Embedded Devices

Official implementation of the paper KairosAD: A SAM-Based Model for Industrial Anomaly Detection on Embedded Devices accepted at the 23rd International Conference on Image Analysis and Processing (ICIAP 2025).

Installation

1. Repository setup:

• $ git clone --recurse-submodules https://github.com/intelligolabs/KairosAD

Or, if you have already has already cloned the repo:

• $ git submodule update --init --recursive
• $ cd KairosAD
• On file MobileSAM/mobile_sam/utils/transforms.py, update the function apply_image() with the following code:

    def apply_image(self, image: np.ndarray) -> np.ndarray:
        """
        Expects a numpy array with shape HxWxC in uint8 format.
        """
        target_size = self.get_preprocess_shape(image.shape[0], image.shape[1], self.target_length)
        image = image.permute(2, 0, 1)
        
        return resize(image, target_size)

• Download the MVTec AD dataset from https://www.mvtec.com/company/research/datasets/mvtec-ad/downloads
• Download the FACES dataset from https://github.com/amazon-science/spot-diff

2. Conda environment setup:

• $ conda create -n kairos_ad python=3.10
• $ conda activate kairos_ad
• $ cd MobileSAM
• $ pip install -e .
• $ pip install -r requirements.txt

Optionally, you can also log the training and evaluation to wandb.

• Update line 102 of the file main.py, specifying project='' and entity=''

Authors

Uzair Khan, Franco Fummi, Luigi Capogrosso

Department of Engineering for Innovation Medicine, University of Verona, Italy

name.surname@univr.it

Citation

If you use KairosAD, please, cite the following paper:

@Article{khan2025kairosad,
  title   = {{KairosAD: A SAM-Based Model for Industrial Anomaly Detection on Embedded Devices}},
  author  = {Khan, Uzair and Fummi, Franco and Capogrosso, Luigi},
  journal = {arXiv preprint arXiv:2505.24334},
  year    = {2025}
}