Neelu Madan, Nicolae-Cătălin Ristea, Radu Tudor Ionescu, Kamal Nasrollahi, Fahad Shahbaz Khan, Thomas B. Moeslund, Mubarak Shah. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024.
In this article, we propose to integrate the reconstruction-based functionality into a novel self-supervised masked convolutional transformer block. The proposed self-supervised block is generic and can easily be incorporated into various state-of-the-art anomaly detection methods.
The official paper can be found at: https://doi.org/10.1109/TPAMI.2023.3322604
The open-access preprint can be found at: http://arxiv.org/abs/2209.12148
This code is released under the CC BY-NC-SA 4.0 license.
Our kernel is illustrated in the picture below. The visible area of the receptive field is denoted by the regions Ki, ∀i ∈ {1, 2, 3, 4}, while the masked area is denoted by M. A dilation factor d controls the local or global nature of the visible information with respect to M.
We provide implementation for both PyTorch and Tensorflow in the torch_ssmctb.py, torch_ssmctb_3d.py and tf_ssmctb.py scripts.
In order to work properly, you need to have a python version newer than 3.6 (we used the python 3.6.8 version).
The Thermal Rare Event dataset is available for download at: https://www.kaggle.com/datasets/neelu1/thermal-anomaly-detection-dataset
It is released under the Attribution 4.0 International (CC BY 4.0) license.
If you use our block in your own work, please don't forget to cite us:
@article{Madan-TPAMI-2024,
title="{Self-Supervised Masked Convolutional Transformer Block for Anomaly Detection}",
author={Madan, Neelu and Ristea, Nicolae-Catalin and Ionescu, Radu Tudor and Nasrollahi, Kamal and Khan, Fahad Shahbaz and Moeslund, Thomas B and Shah, Mubarak},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
year={2024},
volume={46},
number={1},
pages={525--542},
doi={10.1109/TPAMI.2023.3322604}
}
You can send your questions or suggestions to: