Learning Event-Relevant Factors for Video Anomaly Detection

Authors: Che Sun, Chenrui Shi, Yunde Jia, Yuwei Wu

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The extensive experiments show the effectiveness of our method for video anomaly detection.
Researcher Affiliation Academia 1Beijing Key Laboratory of Intelligent Information Technology, School of Computer Science & Technology, Beijing Institute of Technology, China 2Guangdong Laboratory of Machine Perception and Intelligent Computing, Shenzhen MSU-BIT University, China
Pseudocode No The paper does not include pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We conduct experiments on three common benchmark datasets, including Shanghai Tech (Luo, Liu, and Gao 2017), CUHK Avenue (Lu, Shi, and Jia 2013), and UCSD Ped2 (Mahadevan et al. 2009).
Dataset Splits No The paper specifies training and testing scenarios ('Setting-A (big data scenarios): ...all training samples are used for training.', 'Setting-B (small data scenarios): ...We randomly select 10% of the training samples to form sub-datasets...', 'tested on the whole test dataset') but does not explicitly mention a separate validation split or dataset.
Hardware Specification Yes The experiment results are obtained on a single NVIDIA RTX3090 GPU and an Intel i9-10900X CPU, and we do not consider the pre-processing time of the object detection and optical flow estimation.
Software Dependencies No We use Py Torch (Paszke et al. 2017) to train our model and adopt the Adam optimizer (Kingma and Ba 2015) with β1 = 0.9 and β2 = 0.999 to optimize it. While PyTorch is mentioned, a specific version number is not provided.
Experiment Setup Yes The batch size, epoch number and initialized learning rate are set to (128, 80, 1e-4) and (128, 40, 8e-5) for training the causal generative model and finetuning the predictor, respectively. The learning rate is decayed by 0.8 after every 40 epochs. The margin parameter ϵ is set to 1 in the Shanghai Tech dataset and is set to 0.5 in the CUHK Avenue and UCSD Ped2 datasets. λce is a trade-off parameter and is set to 0.001. λdis is a trade-off parameter and is set to 0.5.